Agent Harness Wiki

57 min read

Wiki Log

Chronological record of all wiki actions. Append-only. Format: ## [YYYY-MM-DD] action | subject Actions: ingest, update, query, lint, create, archive, delete When this file exceeds 500 entries, rotate: rename to log-YYYY.md, start fresh.

[2026-04-07] create | Wiki initialized

  • Domain: Agent harnesses — software infrastructure for reliable autonomous coding agents
  • Focus: Codex CLI, Claude Code, Hermes Agent, Gas Town/Gas City, and architectural patterns
  • Structure created with SCHEMA.md, index.md, log.md, raw/, entities/, concepts/, comparisons/, queries/

[2026-04-07] ingest | Bulk ingestion: 15 sources

[2026-04-07] update | Wiki recovery after Codex-assisted audit

[2026-04-07] create | safety-and-permissions

[2026-04-07] create | evaluation-and-review-loops

[2026-04-07] ingest | Current-source refresh from parallel research agents

[2026-04-07] create | codex-app-server

  • Added codex-app-server.md because the App Server is now central enough in OpenAI’s public materials to warrant its own page.
  • Updated index.md to include the new entity and bump total pages from 16 to 17.

[2026-04-07] create | harness-decision-matrix

[2026-04-07] create | automation-and-background-work

[2026-04-07] create | instruction-layering

[2026-04-07] ingest | orchestration and automation docs refresh

[2026-04-07] create | orchestration-topologies

[2026-04-08] ingest | arXiv round two on formal semantics for agent harnesses

[2026-04-08] ingest | formal-core cognition research pass

[2026-04-08] ingest | non-linear harness interface research pass

[2026-04-08] update | add diagrams to core concept pages

[2026-04-08] lint | wiki grooming pass

  • Audited the corpus against the current schema: index coverage, source existence, tag taxonomy, outbound wikilinks, and modified-page updated dates.
  • Fixed the taxonomy violation in gas-city-but-its-just-codex.md by replacing ad hoc tags with existing schema tags.
  • Bumped stale updated dates on the already edited comparison, concept, and query pages so frontmatter matches the current worktree.
  • Added scripts/lint-wiki.sh and documented it in SCHEMA.md so future grooming does not depend on chat memory.

[2026-04-09] ingest | non-hierarchical orchestration research pass

[2026-04-09] ingest | Memento-Skills paper and companion code

[2026-04-09] ingest | broader web patterns for non-linear harness interfaces

[2026-04-09] ingest | under-explored coordination strategies arXiv pass

[2026-04-09] update | page-count recovery note

[2026-04-09] ingest | legacy distributed-systems ideas and studio architecture spec

[2026-04-09] lint | maintenance pass

[2026-04-09] update | staged lint polish

  • Adjusted scripts/lint-wiki.sh so the updated: check inspects staged modified content pages when a commit is in flight, rather than only unstaged worktree edits.
  • Limited that updated: rule to modified tracked pages (--diff-filter=M), so newly introduced historical pages can keep their original creation/update dates.
  • Guarded the git-diff branch so staged-tree audits outside a git checkout do not emit misleading repo warnings.

[2026-04-09] ingest | schema pass and sci-fi audit for moldable operations studio

[2026-04-09] ingest | grounded research for moldable operations studio ideas

[2026-04-09] ingest | wireframe pass for moldable operations studio

[2026-04-09] ingest | self-evolving workflows for gas-city-but-its-just-codex

[2026-04-09] ingest | broader arxiv sweep for self-evolving workflows

[2026-04-09] ingest | rl gyms and executable environments for ai harnesses

[2026-04-10] ingest | neural-native programming via internal-layer interfaces

[2026-04-10] lint | neural-native programming ingest maintenance pass

[2026-04-10] create | neural-native programming research program

[2026-04-10] lint | neural-native research-program follow-up

  • Re-ran scripts/lint-wiki.sh after adding the research-program page.
  • Result: pass, 48 content pages checked.

[2026-04-10] ingest | rl gym entity pages and atropos fit analysis

[2026-04-10] ingest | concrete another-harness atropos schema

[2026-04-10] update | public repo rename and github pages publishing layer

  • Renamed the public GitHub repository from wiki to agent-harness-wiki.
  • Added a lightweight MkDocs publishing layer with:
    • mkdocs.yml
    • docs-requirements.txt
    • scripts/prepare_site_docs.py
    • .github/workflows/pages.yml
    • README.md updates for local preview and published-site usage
    • .gitignore rules for generated site directories and local virtualenvs
  • Enabled GitHub Pages in workflow-deployment mode and updated the Pages workflow to current action major versions to avoid stale Node 20 warnings.
  • Verified the docs build locally after preparing the generated docs tree.
  • Re-ran lint after the publishing-layer pass: pass, 70 content pages checked.

[2026-04-10] update | another-harness work-item closure environment prototype

[2026-04-10] update | seeded user evolution log

  • Added top-level meta file:
    • USER_EVOLUTION_LOG.md
  • Designed a durable entry format that separates explicit observations from inferred trends, records confidence, and preserves evidence, uncertainty, and what-to-watch-next.
  • Seeded initial entries for 2026-04-05 through 2026-04-10 covering:
    • tightening acceptance and review standards
    • shift from compatibility analysis toward switchable Hermes profiles
    • movement toward non-linear / moldable operations studio interfaces
    • clearer separation of formulas, skills, memory, and evaluation in self-evolving workflow research
    • explicit agency-preserving collaboration norms

[2026-04-10] ingest | multiplayer harness and p2p network research pass

[2026-04-10] create | multiplayer harness build strategy

[2026-04-10] create | multiplayer harness inspection guide

[2026-04-10] update | another-harness evaluator-discipline environment prototype

[2026-04-10] ingest | sovereignty and observed-goals deep-dive for multiplayer harnesses

[2026-04-10] create | sovereign identity and observed-goals schema pass

[2026-04-10] create | node card and minimum adapter contract

[2026-04-11] create | prompt optimization research and dspy follow-ups

[2026-04-11] create | prompt optimization timeline and harness lessons

[2026-04-11] create | prompt-program deployment open questions

  • Created 1 new synthesis page focused on deployment-time research gaps for prompt programs and writable instruction artifacts:
  • Updated navigation to include the new page in the query index:
  • Focus: compile-time vs runtime adaptation, promotion and rollback, safety constraints, drift, human oversight, memory substrates, and long-lived harness packaging.
  • Updated index total from 84 to 85.

[2026-04-11] create | prompt optimization open questions fan-out

[2026-04-11] update | another-harness run-history decision

  • Added 1 raw repository-synthesis note:
  • Updated the main Atropos fit synthesis page so it now records the repo’s explicit storage answer:
  • Updated index.md so the query summary reflects the new run-history stance while keeping total content pages at 88.
  • Captured the repo’s current decision that state/runs/ remains non-canonical and that historically important attempts should live as derived evidence bundles under docs/plans/artifacts/ until real multi-attempt pressure says otherwise.

[2026-04-11] lint | wiki pass after another-harness run-history sync

  • Ran scripts/lint-wiki.sh after the run-history decision sync.
  • Confirmed the updated query/index metadata and the new raw note pass the current wiki structural checks.

[2026-04-11] create | self-evolving workflow tool coverage

[2026-04-11] lint | self-evolving workflow tool coverage pass

  • Ran scripts/lint-wiki.sh after adding the workflow, memory, skill-library, and self-evolving-agent entity pages.
  • Confirmed index coverage and wikilink resolution for the new entity set: pass, 113 content pages checked.

[2026-04-11] create | prompt optimization tool coverage

[2026-04-11] create | research on the ten prompt-optimization questions

[2026-04-11] create | another-harness model-docs drift checker

[2026-04-11] lint | wiki pass after model-docs drift sync

  • Ran scripts/lint-wiki.sh after the 0032 sync.
  • Confirmed the new raw note, query page, and index update pass the current wiki structural checks.

[2026-04-11] create | another-harness resume-recover environment

[2026-04-11] lint | wiki pass after resume-recover sync

  • Ran scripts/lint-wiki.sh after the 0036 sync.
  • Confirmed the new raw note, query page, and index update pass the current wiki structural checks.

[2026-04-11] create | three more prompt-optimizer branches

[2026-04-11] update | quartz publishing migration

  • Replaced the GitHub Pages publishing path from MkDocs with Quartz 4 while keeping the repository-root markdown corpus canonical.
  • Added scripts/prepare_quartz_content.py to materialize the wiki into .quartz-site/content/, synthesize titles for root meta pages, and sanitize frontmatter for Quartz consumption.
  • Vendored Quartz under .quartz-site/ from upstream commit 59b5807, customized the local Quartz config/layout, and switched .github/workflows/pages.yml to a Node 22 + Quartz build.
  • Verified the Quartz build locally; the explorer/sidebar navigation replaces the brittle top dropdown navbar, and Mermaid diagrams render with built-in expand/zoom controls.

[2026-04-11] create | repo architecture plans from the ten prompt-program questions

  • Created 2 repo-grounded plan documents that translate the ten cross-cutting prompt-program questions into concrete architecture stances and sequencing ladders:
    • /Users/ericfode/src/another-harness/docs/plans/2026-04-11-another-harness-prompt-program-architecture-plan.md
    • /Users/ericfode/src/gas-city-but-its-just-codex/docs/plans/2026-04-11-gascity-prompt-program-architecture-plan.md
  • Added 1 wiki synthesis page summarizing both repo plans and their main divergence in design center:
  • Updated prompt-program navigation so the new cross-repo plan page is indexed from the query catalog and linked from the umbrella question map:
  • Updated index total from 126 to 127.

[2026-04-11] create | context visualization for harnesses

[2026-04-12] update | repo recovery and scratch hygiene

  • Recovered the stranded local wiki additions from the older spike/quartz-nav-rescue worktree onto main, including the resume-recover and context-visualization query pages, their supporting raw notes, and one additional workflow-visualization research note.
  • Updated context-engineering.md and web-patterns-for-non-linear-harness-interfaces.md so the new research is connected into the existing graph, and normalized index.md to 130 pages.
  • Added repo ignore rules for local scratch directories and moved the generated Quartz content tree to local git excludes so builds stop cluttering status without hiding the content from Quartz itself.

[2026-04-12] lint | recovery pass

  • Ran scripts/lint-wiki.sh after the recovery merge and ignore cleanup.
  • Result: pass, 130 content pages checked.

[2026-04-14] ingest | mathcode

[2026-04-14] lint | mathcode ingest

  • Ran scripts/lint-wiki.sh after adding MathCode to the wiki.
  • Result: pass, 131 content pages checked.

[2026-04-14] update | wiki maintenance pass

[2026-04-14] lint | wiki maintenance pass

  • Ran scripts/lint-wiki.sh after the repair pass.
  • Ran an additional cross-link audit for broken wikilinks and orphaned content pages.
  • Result: pass, 131 content pages checked, 0 broken wikilinks, 0 orphan pages.

[2026-04-15] lint | pre-commit wiki rerun

  • Re-ran scripts/lint-wiki.sh on the current wiki worktree before commit and push.
  • Normalized updated frontmatter on the previously touched query/entity/concept pages from 2026-04-14 to 2026-04-15 so the modified-page freshness rule reflects the actual edit date.
  • Result: pass, 131 content pages checked.

[2026-04-15] update | gas-city-but-its-just-codex current-structure deep dive

  • Added 1 fresh raw repository snapshot grounded in the live worktree rather than only the last clean commit:
  • Rewrote gas-city-but-its-just-codex.md from an early-spike design note into an up-to-date structure tour covering the ledger, formulas, app service, control surfaces, operator daemon, sidecar/UI path, operational center of gravity, and current structural tensions.
  • Updated index.md so the query summary and index header date reflect the refreshed deep dive.

[2026-04-15] lint | gas-city-but-its-just-codex deep dive refresh

  • Ran scripts/lint-wiki.sh after refreshing the raw note, query page, and index/log metadata.
  • Also re-checked broken wikilinks and orphaned content pages after the rewrite.
  • Result: pass, 131 content pages checked, 0 broken wikilinks, 0 orphan pages.

[2026-04-15] update | gas-city-but-its-just-codex documentation rendering pass

[2026-04-15] lint | gas-city documentation rendering pass

  • Ran scripts/lint-wiki.sh after adding the new raw note, three focused query pages, and the navigation updates.
  • Also re-checked broken wikilinks and orphaned content pages after the rendering pass.
  • Result: pass, 134 content pages checked, 0 broken wikilinks, 0 orphan pages.

[2026-04-21] update | nightly src projects desk kickoff

  • Added 1 raw survey note grounding a public-safe snapshot of the local /Users/ericfode/src tree:
  • Added 1 new query page for the first nightly project-news brief:
  • Added a generated editorial illustration under queries/news-assets/ and wired it into the new desk page.
  • Updated index.md with a new News section and a corresponding query listing while bumping the header date/page count.

[2026-04-21] update | nightly journalist loop automation

  • Created a nightly Hermes cron job to resurvey /Users/ericfode/src, refresh the public-safe project desk, generate fresh editorial art, run wiki lint, commit the new desk update, and push the wiki repository.
  • The loop is configured to use 10 survey lanes over the src/ tree, with recursive subagent fanout requested where the runtime exposes it and a documented parallel-inspection fallback when it does not.

[2026-04-21] lint | nightly src projects desk kickoff

  • Ran scripts/lint-wiki.sh after adding the raw survey note, the new desk page, the illustration asset, and the index/log updates.
  • Result: pass, 136 content pages checked.

[2026-05-01] create | software verification/testing environment project anchor

[2026-05-01] lint | software verification/testing environment project anchor

  • Ran scripts/lint-wiki.sh after creating the project anchor page and inbound links.
  • Result: pass, 137 content pages checked.

[2026-05-01] create | agent-facing-verifier-environment-architecture

[2026-05-01] update | agent-facing verifier architecture — t_2cf92f9b repairs

[2026-05-01] ingest | software verification/testing environment research synthesis

[2026-05-01] lint | verify: lint and repair wiki after verification/testing project update

[2026-05-02] query | nightly src projects desk

  • Created raw survey note nightly-src-projects-desk-2026-05-02.md and public query page nightly-src-projects-desk-2026-05-02.md.
  • Generated local editorial SVG illustration at queries/news-assets/2026-05-02-project-desk-hero.svg after the configured image backend reported FAL_KEY was unavailable; no fake screenshots were used.
  • Survey automation status: exactly 10 top-level Hermes survey lanes covered all 27 top-level directories under /Users/ericfode/src; each lane reported deeper three-way delegation was available and used. A first all-at-once 10-task batch was rejected by the runtime concurrency cap, so the lanes ran as two batches of five.
  • Public-safety filter: project-specific details were held back for 10 surveyed directories because of sensitive identity/reputational material, mixed creative material needing curation, credential/private deployment signals, internal supervisor/evaluator material, scratch wrappers, empty directories, or skeletal non-project evidence.
  • Updated index.md News and Queries sections for the new nightly desk page; working-tree page count moved 140 → 141.
  • Lint result: initial scripts/lint-wiki.sh run failed on 21 pre-existing dirty content pages whose updated fields were still 2026-05-01 under a 2026-05-02 clock. After staging only the intended nightly desk content page so the linter used the cached content-page set for modified-page freshness, scripts/lint-wiki.sh passed: 141 content pages checked.

[2026-05-02] lint | verification/testing project freshness repair

[2026-05-02] query | specification elaboration naming frame

  • Added raw research note specification-elaboration-naming-research.md covering knowledge representation, formal concept analysis, Problem Frames, Moldable Development, and knowledge maps as naming constraints.
  • Created specification-elaboration-naming-frame.md.
  • Recommendation recorded: replace “spec image” with Intent Lattice as the working product frame, with intent model, intent graph, viability critique, refinement packet, and adapter as internal vocabulary.
  • Updated index.md to include the new query page and bump total pages from 141 to 142.

[2026-05-02] query update | specification elaboration naming frame

  • Incorporated the user’s clarification that the project is about making specifications more verifiable, not imaging them.
  • Added NASA requirements-verifiability guidance, SpecRL negative-test reward framing, and LTL specification-mining context to specification-elaboration-naming-research.md.
  • Revised specification-elaboration-naming-frame.md: Claim Lattice is now the working product/kernel name, and collaborative spec verifiability environment is the category phrase.
  • Added the future RL-environment boundary: observation, action, actor, reward, done, and safety invariant.
  • Added the human-play boundary: the human is a first-class actor whose refinements, overrides, rejections, and rationale are durable environment transitions.
  • Updated index.md to reflect the revised frame.

[2026-05-02] query update | process-first CLI naming

  • Incorporated the user’s correction that Claim Lattice names the internal model, not the process, and is therefore a poor CLI name.
  • Revised specification-elaboration-naming-frame.md: Spec Gym / specgym is now the recommended CLI/product name; Claim Lattice remains the internal model.
  • Added command vocabulary: play, score, step, rollout, and export.
  • Added practical npm registry checks in specification-elaboration-naming-research.md: specgym, spec-gym, speclab, spec-lab, proveout, spec-assay, and spec-trial returned 404; specforge, assay, sharpen, and spec-check already exist.
  • Updated index.md to use the specgym frame.

[2026-05-02] decision | Spec Gym accepted

[2026-05-02] implementation | Spec Gym repo vocabulary migration

  • Propagated the accepted Spec Gym / specgym name into /Users/ericfode/Documents/New project 4.
  • The top-level repo spec now describes a collaborative spec verifiability environment, not a spec-imaging tool.
  • Primary artifacts are claim-lattice.json, viability-critique.md, ClaimLattice.lean, and refinement-packet.md.
  • The generated claim lattice now includes environment fields for observation, action, actor, reward signals, done states, safety invariant, and score.
  • Linear project renamed to Spec Gym; FOD-30 marked done after local verification.

[2026-05-02] ingest | 0xSero self-distillation video

[2026-05-03] query | nightly src projects desk

  • Created raw survey note nightly-src-projects-desk-2026-05-03.md and public query page nightly-src-projects-desk-2026-05-03.md.
  • Generated editorial illustration queries/news-assets/2026-05-03-project-desk-hero.svg locally after the configured image backend reported missing FAL_KEY; no fake screenshots were used.
  • Survey automation status: exactly 10 top-level Hermes survey lanes covered all 30 top-level directories under /Users/ericfode/src; a first 10-lane batch was rejected by the runtime concurrency cap (max_concurrent_children=6), so the lanes ran as 6 + 4. Lane summaries reported orchestrator-style project splits and focused purpose/docs, live-work, and safety passes, with one project child handled directly rather than recursively.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for 10 surveyed directories because of sensitive identity/reputational framing, local agent settings, empty/skeletal evidence, credential/private deployment signals, uncommitted internal workbenches, mixed creative material needing curation, hidden evaluator/supervisor surfaces, or prompt-bearing workflow residue.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 143 → 144.
  • Lint result: scripts/lint-wiki.sh passed: 144 content pages checked.

[2026-05-04] query | LLM-readable spec files

  • Created raw source note llm-readable-spec-files-research-2026-05-04.md covering OpenAI prompting/evals guidance, Claude Code memory rules, GitHub Copilot instructions, GitHub Spec Kit, NASA/INCOSE requirements guidance, Microsoft LLM-judge reliability notes, and nl2postcond.
  • Created query page llm-readable-spec-files.md.
  • Main conclusion: excellent spec.md files are compact executable contracts with stable IDs, glossary, scope, non-goals, behavioral requirements, invariants, examples/counterexamples, verification matrix, operational gates, and explicit unresolved questions.
  • Updated index.md to include the new query page and bump total pages from 144 to 145.

[2026-05-04] query | spec dataset evolution research project

  • Created Kanban board spec-dataset-evolution for the public spec.md / spec-like corpus project.
  • Created raw design note spec-dataset-evolution-kanban-design-2026-05-04.md from three methodology lanes: corpus discovery, longitudinal change analysis, and code/spec connectedness.
  • Created query page spec-dataset-evolution-research-project.md.
  • Added scratch project brief at /Users/ericfode/.hermes/gateway-scratch/spec-dataset-evolution/PROJECT_BRIEF.md.
  • Kanban tasks created: t_39d3ad17 through t_d29a8854; anchor task t_39d3ad17 marked done, with four methodology tasks now ready.
  • Updated index.md to include the new query page and set content-page count to 146.
  • Lint result: scripts/lint-wiki.sh passed: 146 content pages checked.

[2026-05-04] update | spec dataset evolution deep-repo expansion

  • User clarified that the project should fan out more and deeply explore whole repositories, not merely design a file-level crawler.
  • Added scratch protocol DEEP_REPO_EXPLORATION_PROTOCOL.md requiring each scout to inspect 3–5 public repos, including whole-repo structure, spec inventory, git history, code/spec ratios, connectedness, pressure signals, compliance, and structured dossier artifacts.
  • Added Kanban tasks SPEC-REPO-00 through SPEC-REPO-14; SPEC-REPO-00 / t_2d3d9d2f was completed as the protocol anchor, and 12 scout tasks plus aggregation/review tasks were created.
  • Linked SPEC-REPO-14 / t_482593d1 as a parent of SPEC-DATA-13 / t_b5c6851b, so the pilot report cannot complete without the deep-repo evidence review.
  • Updated spec-dataset-evolution-research-project.md to record the deep-repo wave and task IDs.

[2026-05-05] update | spec dataset evolution Kanban error repair

  • Mirrored scratch repair note into the wiki as spec-dataset-evolution-kanban-error-repair-2026-05-05.md.
  • Updated spec-dataset-evolution-research-project.md sources and operational repair notes.
  • Recorded root cause: default Kanban workers crash-looped because the globally enabled Basis plugin used top-level anyOf in tool parameter schemas rejected by OpenAI Codex.
  • Recorded repair: local Basis plugin source commit 0061d32 made the schemas Codex-compatible; SPEC-DATA-JCODE completed, SPEC-DATA-21 was manually closed after verified artifacts, and SPEC-DATA-22 intentionally blocked on unresolved human/adjudicated audit labels.

[2026-05-05] update | private spec corpus and wiki-ingest project

  • Created and pushed private GitHub repository ericfode/spec-dataset-evolution-corpus for the spec dataset pilot corpus.
  • The private corpus includes 1,741 artifact occurrence rows, 1,676 private raw-file copies, provenance manifests, dedup outputs, authored deep-dive dossiers, and validation scripts; 65 hard-quarantined raw rows remain metadata-only.
  • Created Hermes Kanban board spec-deep-dive-wiki-ingest with tasks SPEC-WIKI-00 through SPEC-WIKI-06 to ingest public-safe deep dives into the wiki.
  • Added spec-deep-dive-wiki-ingest-project and linked it from spec-dataset-evolution-research-project.

[2026-05-05] update | spec deep-dive source map and safety policy

  • Expanded spec-deep-dive-wiki-ingest-project with the SPEC-WIKI-00 source map, downstream task source routing, naming conventions, citation style, excerpt policy, fail-closed public-safety gates, and downstream page acceptance criteria.
  • Wrote machine-readable handoff artifact source-map-public-safety-policy.json in the project scratch directory for downstream workers.
  • Source evidence inspected: private corpus checkout HEAD 4659608, clean working tree, 207 deep-dive files, 1,741 manifest rows, 55 aggregate repo records, and 55 connectedness feature rows.
  • Validation: scripts/lint-wiki.sh passed: 147 content pages checked.

[2026-05-05] query | spec deep-dive priority case studies

  • Created three public-safe case pages from the private corpus priority dossiers:
    • spec-deep-dive-case-jcode — 1jehuang/jcode as a post-LLM coding-agent harness with distributed spec surfaces and high code/spec connectedness.
    • spec-deep-dive-case-droidagent — DroidAgent as agent-generated mobile-GUI behavioral specs connected to traces, reports, and replay scripts.
    • spec-deep-dive-case-j8-ambiguity — corrected J8/J8Spec ambiguity trail preserving negative evidence and the pre-AI J8Spec executable-spec control.
  • Updated spec-deep-dive-wiki-ingest-project, spec-dataset-evolution-research-project, and index for inbound links and page count.
  • Source evidence used: reports/deep-dives/jcode.md, reports/deep-dives/jcode-analysis.json, reports/jcode_first_calibration_seed.md, reports/deep-dives/droidagent.md, and reports/deep-dives/j8-agent.md from the private corpus checkout at HEAD 4659608.
  • Public-safety boundary: synthesis and metadata only; no raw private corpus bodies or long upstream excerpts copied into the wiki.
  • Validation: scripts/lint-wiki.sh passed: 153 content pages checked. During lint repair, index was reconciled with sibling cohort pages already present in the shared worktree, and the agent-native cohort frontmatter was normalized to existing wiki source pages.

[2026-05-05] query | spec deep-dive agent-native / Spec Kit / Kiro cohort

  • Created spec-deep-dive-cohort-agent-native-spec-kit-kiro from private corpus cohorts SPEC-REPO-03, SPEC-REPO-04, SPEC-REPO-08, SPEC-REPO-11, and SPEC-REPO-20 template-lineage reports.
  • Preserved the template-vs-independent-spec distinction: 19 relevant selected dossier occurrences, 15 generated-scaffold rows, 11 edited-project-contract rows, and 14 fail-closed independent-authority weights in the lineage mapping.
  • Source evidence used: reports/deep-dives/SPEC-REPO-03/index.md, reports/deep-dives/SPEC-REPO-04/index.md, reports/deep-dives/SPEC-REPO-08/index.md, reports/deep-dives/SPEC-REPO-11/index.md, and reports/deep-dives/SPEC-REPO-20/TEMPLATE_LINEAGE_MODEL.md from the private corpus checkout at HEAD 4659608.
  • Public-safety boundary: synthesis and metadata only; no raw private corpus bodies or long upstream excerpts copied into the wiki.

[2026-05-05] query | spec deep-dive RFC, ADR, and executable-contract cohort

[2026-05-05] query | spec deep-dive exact spec.md and standards cohort

  • Created spec-deep-dive-cohort-exact-spec-md-and-standards from private corpus cohorts SPEC-REPO-01, SPEC-REPO-07, and related standards records in SPEC-REPO-18 and SPEC-REPO-19.
  • Distinguished exact lower-case spec.md as a discovery handle from repository authority shape: OCI runtime/distribution, Compose Spec, HCL, and Cloud Foundry Service Broker distribute authority across prose, schemas, conformance tests, implementation code, profile material, release tags, and governance.
  • Mature baseline evidence from OCI runtime, OpenTelemetry, Protocol Buffers, and gRPC was summarized as metadata and synthesis only; no private raw corpus bodies or long upstream excerpts were copied into the wiki.
  • Updated index and reused existing cross-links from spec-deep-dive-wiki-ingest-project and spec-dataset-evolution-research-project so the new cohort page participates in the shared ingest graph.
  • Validation: scripts/lint-wiki.sh passed: 153 content pages checked.

[2026-05-05] query | spec deep-dive aggregate index

  • Created spec-deep-dive-index as the public-safe aggregate index for the private spec corpus deep-dive wave.
  • Cross-linked the aggregate index from spec-dataset-evolution-research-project, spec-deep-dive-wiki-ingest-project, index, and the priority case/cohort pages.
  • Preserved the private corpus as a source pointer only: aggregate counts, corpus-relative source paths, repo URLs, connectedness/compliance summaries, and caveats are public; raw private corpus bodies remain unpublished.
  • Validation: scripts/lint-wiki.sh passed: 154 content pages checked.

[2026-05-05] query | hidden-state JEPA reasoning shortcuts

[2026-05-06] query | nightly src projects desk

  • Created raw survey note nightly-src-projects-desk-2026-05-06.md and query page nightly-src-projects-desk-2026-05-06 for the nightly local-projects desk.
  • Published verdict: Basis/spec-code grounding and test-writing environments lead tonight, with tinygrad/Gemma benches, handterm, Dungeon Steward, FACEMUSIC, orchestration side rooms, and privacy-gated corpus work summarized only where inspectable evidence supported public claims.
  • Generated editorial illustration queries/news-assets/2026-05-06-project-desk-hero.svg locally after the configured image backend reported missing FAL_KEY; no fake screenshots were used.
  • Survey automation status: exactly 10 top-level Hermes survey lanes covered all 38 top-level directories under /Users/ericfode/src; the lanes ran as 6 + 4, and all lane summaries reported successful recursive 3-way delegation for purpose/docs, live-work evidence, and safety/public-summary review.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for 13 surveyed directories because of local hidden settings, sensitive social/reputational framing, empty or skeletal evidence, private corpus contents, local deployment/model-runner configuration, raw logs/prompts/trajectories, evaluator-like materials, generated artifacts, or creative material needing human curation.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 155 → 156.
  • Lint result: scripts/lint-wiki.sh passed: 156 content pages checked.

[2026-05-06] update | 0xSero self-distillation post recheck

  • Rechecked the user-supplied canonical X/Twitter URL https://twitter.com/0xSero/status/2050692154524156374 against public oEmbed metadata.
  • Updated the existing raw source note 0xsero-self-distillation-video-2026-05-02.md rather than creating a duplicate page; the post had already been ingested on 2026-05-02 and linked to on-policy-self-distillation.
  • No index-page count change: this was a raw-source metadata update, not a new content page.

[2026-05-06] ingest | The Last Harness You’ll Ever Build

  • Resolved the user-supplied screenshot to arXiv:2604.21003v3, arxiv-seong-2026-last-harness-youll-ever-build.md, by grounding metadata through the arXiv Atom API and section text through the arXiv HTML page after web_extract reported Firecrawl was not configured.
  • Created last-harness-youll-ever-build as a source-quality-aware entity page with a Mermaid architecture sketch and evidence/caveat table.
  • Updated harness-engineering and self-evolving-workflows so the new paper is linked from the existing harness-engineering graph rather than sitting as a decorative orphan.
  • Updated index.md to list the new entity page; content-page count moved 156 → 157.
  • Validation: scripts/lint-wiki.sh passed: 157 content pages checked.

[2026-05-06] ingest | Agentic-imodels

  • Resolved the user-supplied arXiv URL https://arxiv.org/abs/2605.03808 to arXiv:2605.03808v1, arxiv-singh-2026-agentic-imodels.md, by grounding metadata through the arXiv Atom API and source text through the arXiv e-print package after web_extract reported Firecrawl was not configured and arXiv HTML returned 404.
  • Created agentic-imodels as a source-quality-aware entity page with an autoresearch-loop diagram, evidence/caveat table, BLADE result summary, and reward-hacking boundary.
  • Updated context-engineering, evaluation-and-review-loops, and self-evolving-workflows so agent-facing simulatability tests are linked into the harness graph.
  • Updated index.md to list the new entity page; content-page count moved 157 → 158.
  • Validation: scripts/lint-wiki.sh passed: 158 content pages checked.

[2026-05-07] query | nightly src projects desk

  • Created raw survey note nightly-src-projects-desk-2026-05-07.md and query page nightly-src-projects-desk-2026-05-07 for the nightly local-projects desk.
  • Published verdict: spec-code grounding, test-writing environments, and tinygrad-gemma lead tonight, with terminal/game/interface craft projects, NNPL benches, and orchestration side rooms summarized only where inspectable evidence supported public-safe claims.
  • Generated editorial illustration queries/news-assets/2026-05-07-project-desk-hero.svg locally after the configured image backend reported missing FAL_KEY; no fake screenshots were used, and the SVG parsed as valid XML.
  • Survey automation status: exactly 10 top-level Hermes survey lanes ran as 3 + 3 + 3 + 1, and all 10 lane summaries reported successful recursive 3-way delegation for purpose/docs, live-work evidence, and safety/public-summary review. A controller post-dispatch audit covered two hidden top-level directories as a read-only supplement so the total lane count remained exactly ten; total coverage was 38 top-level directories.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local settings, one sensitive social-claim notebook, empty or skeletal directories, local deployment/model-runner folders, private corpus contents, internal workflow/assistant configuration, scratch/meta workspaces, raw logs/prompts/trajectories, evaluator-like payloads, benchmark outputs, model/checkpoint artifacts, and creative material needing human curation.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 158 → 159.
  • Lint result: scripts/lint-wiki.sh passed: 159 content pages checked.

[2026-05-08] query | nightly src projects desk

  • Created raw survey note nightly-src-projects-desk-2026-05-08.md and query page nightly-src-projects-desk-2026-05-08 for the nightly local-projects desk.
  • Published verdict: spec-code grounding and test-writing environments still lead, with tinygrad-gemma, NNPL benches, handterm, Dungeon Steward, privacy-sensitive interface/simulation work, and orchestration side rooms summarized only where inspectable evidence supported public-safe claims.
  • Generated editorial illustration queries/news-assets/2026-05-08-project-desk-hero.svg locally after the configured image backend reported missing FAL_KEY; no fake screenshots were used, and the SVG parsed as valid XML.
  • Survey automation status: exactly 10 top-level Hermes survey lanes ran as 3 + 3 + 3 + 1 across all 38 top-level directories under the local src tree, including hidden directories. All 10 lane summaries reported successful recursive 3-way delegation for purpose/docs, live-work evidence, and safety/public-summary review.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local settings, a sensitive social-claim notebook, empty or skeletal directories, local deployment/model-runner folders, private corpus bodies, internal workflow/assistant configuration, scratch/meta workspaces, generated artifacts, prompt/log/trajectory materials, evaluator-like payloads, benchmark raw outputs, model/checkpoint artifacts, privacy-sensitive capture data, and creative material needing human curation.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 159 → 160.
  • Lint result: scripts/lint-wiki.sh passed after the page, index, and log edits: 160 content pages checked.

[2026-05-09] query | nightly src projects desk

  • Created raw survey note nightly-src-projects-desk-2026-05-09.md and query page nightly-src-projects-desk-2026-05-09 for the nightly local-projects desk.
  • Published verdict: spec-code grounding and test-generation environments still lead, with tinygrad-gemma, Gemma/tinygrad optimization benches, fresh kettlebellsim simulation work, Dungeon Steward, handterm, privacy-sensitive interface work, NNPL benches, and orchestration side rooms summarized only where inspectable evidence supported public-safe claims.
  • Generated editorial illustration queries/news-assets/2026-05-09-project-desk-hero.svg locally as symbolic SVG art; no fake screenshots were used.
  • Survey automation status: exactly 10 top-level Hermes survey lanes ran as 3 + 3 + 3 + 1 across all 38 top-level directories under the local src tree, including hidden directories. All 10 lane summaries reported three-way delegation for purpose/docs, live-work evidence, and safety/public-summary review, plus a further three-way leaf recursion where the runtime allowed it; further recursion ended at leaf workers because of the configured depth cap.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local settings, a sensitive social-claim wiki, empty or skeletal directories, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator-like payloads, hidden references/oracles, benchmark raw outputs, model/checkpoint artifacts, privacy-sensitive capture data, creative canon drafts, and local service configuration.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 160 → 161.
  • Lint result: scripts/lint-wiki.sh passed after the page, index, and log edits: 161 content pages checked.

[2026-05-10] query | nightly src projects desk

  • Created raw survey note nightly-src-projects-desk-2026-05-10.md and query page nightly-src-projects-desk-2026-05-10 for the nightly local-projects desk.
  • Published verdict: harness/control-plane work, spec-code reducers, test-generation environments, tinygrad-gemma, NNPL boundary experiments, kettlebellsim, handterm, and selected side rooms lead tonight only where inspectable evidence supported public-safe claims.
  • Generated editorial illustration queries/news-assets/2026-05-10-project-desk-hero.svg locally as symbolic SVG art; the SVG parsed as valid XML and no fake screenshots were used.
  • Survey automation status: exactly 10 top-level Hermes survey lane identities ran in one batch across all 38 top-level directories under the local src tree, including hidden directories. All 10 lane summaries reported three-way delegation for purpose/docs, live-work evidence, and safety/public-summary review, plus one further three-way leaf recursion; further recursion ended at the configured depth cap. A controller read-only audit corrected a lane spelling slip for the actual nnpl-* directories while preserving the top-level lane count of exactly 10.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local settings, hidden-only or empty directories, one sensitive social-claim note set, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator-like payloads, hidden references/oracles, benchmark raw outputs, model/checkpoint artifacts, biometric/capture data, creative story/canon drafts, service configuration, and cache/build/vendor directories.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 161 → 162.
  • Lint result: scripts/lint-wiki.sh passed after the page, index, and log edits: 162 content pages checked.

[2026-05-10] query | Basis project corner

  • Created raw public-safe evidence note basis-project-deep-dive-2026-05-10.md from local inspection of basis, basis-hermes, and basis-jcode.
  • Created basis-project-index, basis-architecture-and-plans, basis-experiment-status, and basis-source-basis-and-safety-gate as a dedicated wiki corner for Basis plans, running experiments, status, and publication boundaries.
  • Recorded verified status: core basis clean at a5544e0, tests and compile pass but formatter gate fails on lib/basis/run/server.ex; basis-hermes clean at 0061d32 with Python and JS gates passing; basis-jcode at 4b1e621, ahead 10 and dirty, with 26 reducer tests passing and a loopback dashboard for .basis/self-convergence.
  • Public-safety boundary: .basis run bodies, prompts, NDJSON streams, packet bodies, validation bodies, logs, dashboard payloads, and private corpus bodies remain unpublished; wiki pages use synthesis, counts, safe filenames, test results, caveats, and plans only.

[2026-05-11] query | nightly src projects desk

  • Created raw survey note nightly-src-projects-desk-2026-05-11.md and query page nightly-src-projects-desk-2026-05-11 for the nightly local-projects desk.
  • Published verdict: test-generation environments, Basis/spec-code work, gemma-dungeon, tinygrad-gemma, NNPL boundary experiments, kettlebellsim, handterm, and selected orchestration side rooms lead tonight only where inspectable evidence supported public-safe claims.
  • Generated editorial illustration queries/news-assets/2026-05-11-project-desk-hero.svg locally as symbolic SVG art; the SVG parsed as valid XML and no fake screenshots were used.
  • Survey automation status: exactly 10 top-level Hermes survey lane identities ran in one batch across all 39 top-level directories under the local src tree, including hidden directories. All 10 lane summaries reported three-way delegation for purpose/docs/manifests, live-work evidence, and public-safety review, plus one further three-way leaf recursion; further recursion ended at leaf checks/depth limits. A controller read-only audit re-enumerated the tree and found no missing top-level directory.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local settings, hidden-only or empty directories, one sensitive social-claim notebook, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator-like payloads, hidden references/oracles, benchmark raw outputs, model/checkpoint artifacts, biometric/capture data, creative story/canon drafts, service configuration, raw test/counterexample bodies, cache/build/vendor directories, and too-skeletal placeholders.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 166 → 167.
  • Lint result: scripts/lint-wiki.sh passed after the page, index, and log edits: 167 content pages checked.

[2026-05-11] update | website navigation and missing page routing

  • Added project-list and project-update-log as curated public-safe landing pages for active project lines and personal project update logs.
  • Added root website doors projects.md and news.md, and updated index.md with a Start Here block plus a shorter News section.
  • Path-qualified ambiguous nightly desk links so query pages resolve to queries/* instead of colliding with raw source notes.
  • Added Quartz trailing-slash redirect aliases and a cleaner Explorer configuration so Basis pages, news, and projects are easier to reach from the published site.

[2026-05-11] update | Basis Reduce and Imagine wiki recovery

[2026-05-13] query | nightly src projects desk

  • Created raw survey note nightly-src-projects-desk-2026-05-13.md and query page nightly-src-projects-desk-2026-05-13 for the nightly local-projects desk.
  • Published verdict: testing-rl and tinygrad-gemma lead tonight, with handterm and Dungeon Steward as clean craft/game leads; Basis/Steward, NNPL, gemma-dungeon, openai-symphony, and kettlebellsim are research-bench or side-room material.
  • Generated editorial illustration queries/news-assets/2026-05-13-project-desk-hero.svg locally as symbolic SVG art; the SVG parsed as valid XML and no fake screenshots were used.
  • Survey automation status: exactly 10 top-level Hermes survey lane identities ran in one batch across all 41 top-level directories under the local src tree, including hidden directories. All 10 lanes spawned three read-only subteams for purpose/docs/manifests, live-work evidence, and public-safety review. Subteams generally recursed once more into leaf probes; two minor leaf-shape exceptions were recorded in the raw note. A controller audit found 41 unique assignments, no missing directories, no extras, and no duplicates.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local settings, internal security-scan artifacts, hidden-only or empty directories, one sensitive social-claim notebook, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator-like payloads, hidden references/oracles, benchmark raw outputs, model/checkpoint artifacts, biometric/capture data, creative story/canon drafts, service configuration, raw test/counterexample bodies, cache/build/vendor directories, and too-skeletal placeholders.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 171 → 172. Existing untracked .hermes/ material in the wiki worktree remained untouched and uncommitted.
  • Lint result: scripts/lint-wiki.sh passed after the new page, index, and log edits: 172 content pages checked.

[2026-05-13] update | East Bay property monitor website publishing

  • Added the focused East Bay Catholic property monitor to the Quartz publishing content tree and linked it from index.md Start Here navigation.
  • Published target page: East Bay Catholic Property Sales Monitor under east-bay-catholic-property-monitor/queries/.
  • Preserved the main harness-wiki content count at 172; the property monitor remains a focused sub-vault rather than part of the main harness taxonomy.

[2026-05-14] query | nightly src projects desk

  • Created raw survey note raw note and query page nightly-src-projects-desk-2026-05-14 for the nightly local-projects desk.
  • Published verdict: gemma-dungeon is tonight’s same-night lead, textual-world-model is active benchmark-first research-loop evidence, testing-rl and tinygrad-gemma remain the stable benches, and Basis/Steward/Symphony/Gas-City work stays architecture/provenance-level under the safety filter.
  • Generated editorial illustration queries/news-assets/2026-05-14-project-desk-hero.svg as deterministic symbolic SVG after rejecting a raster draft with text artifacts; no fake screenshots were used, and the SVG parsed as valid XML.
  • Survey automation status: exactly 10 top-level Hermes survey lane identities ran in one batch across all 41 top-level directories under the local src tree, including hidden directories. All 10 lanes reported three read-only subteams for purpose/docs/manifests, live-work evidence, and public-safety review, plus one further three-way leaf recursion where delegation was exposed. A controller audit found 41 unique assignments, no missing directories, no extras, and no duplicates.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local settings, internal security-scan artifacts, empty or hidden-only directories, one protected-class-sensitive social-claim notebook, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator/oracle payloads, benchmark raw outputs, model/checkpoint artifacts, biometric/capture data, creative/canon drafts, service configuration, raw test/counterexample bodies, cache/build/vendor directories, and too-skeletal placeholders.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 172 → 173. Existing untracked .hermes/ material in the wiki worktree remained untouched and uncommitted.
  • Lint result: bash scripts/lint-wiki.sh passed after the new page, index, and log edits: 173 content pages checked.

[2026-05-16] query | nightly src projects desk

  • Created raw survey note raw note and query page nightly-src-projects-desk-2026-05-16 for the nightly local-projects desk.
  • Published verdict: jepa-poker is tonight’s fresh public-safe lead, with gemma-dungeon, testing-rl, and tinygrad-gemma as continuing benches; is-it-formal and basis-hermes give the formal/spec-provenance corner clean surfaces; handterm, Dungeon Steward, and kettlebellsim remain craft/game/simulation side rooms.
  • Generated editorial illustration queries/news-assets/2026-05-16-project-desk-hero.svg as deterministic symbolic SVG after rejecting a generated raster draft with text artifacts; no fake screenshots were used, and the SVG parsed as valid XML.
  • Survey automation status: exactly 10 top-level Hermes survey lane identities ran in one batch across all 43 top-level directories under the local src tree, including hidden directories. All 10 lanes reported three read-only subteams for purpose/docs/manifests, live-work evidence, and public-safety review, plus one further three-way leaf recursion. A controller audit found 43 assigned directories, 43 unique assignments, no missing directories, no extras, and no duplicates.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local settings, security-scan artifacts, empty or hidden-only directories, provocative/protected-class-sensitive social-claim material, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator/oracle payloads, benchmark raw outputs, model/checkpoint artifacts, biometric/capture data, creative/canon drafts, service configuration, raw test/counterexample bodies, cache/build/vendor directories, and too-skeletal placeholders.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 173 → 174. Existing unrelated East Bay property-monitor working-tree changes and wiki .hermes/ material remained untouched and uncommitted.
  • Lint result: bash scripts/lint-wiki.sh passed after the new page, index, and log edits: 174 content pages checked.

[2026-05-24] query | nightly src projects desk

  • Created raw survey note raw note and query page nightly-src-projects-desk-2026-05-24 for the nightly local-projects desk.
  • Published verdict: unconventional-jepa-lab is tonight’s clean public lead; word-games and jepa-lang make the JEPA/world-model line newly legible; testing-rl, tinygrad-gemma, and gemma-dungeon remain the sturdy continuing benches; Basis/Hermes, Lean/formality, and Steward keep the formal/spec/provenance corner coherent.
  • Generated editorial illustration queries/news-assets/2026-05-24-project-desk-hero.svg as deterministic symbolic SVG; no fake screenshots were used, and the SVG parsed as valid XML.
  • Survey automation status: exactly 10 top-level Hermes survey lane identities ran as two batches of five across all 47 top-level directories under the local src tree, including hidden directories. All 10 lanes reported three read-only subteams for purpose/docs/manifests, live-work evidence, and public-safety review, plus one further three-way leaf recursion. A controller audit found 47 assigned directories, 47 unique assignments, no missing directories, no extras, and no duplicates.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local settings, security-scan artifacts, empty or hidden-only directories, provocative/protected-class-sensitive social-claim material, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator/oracle payloads, benchmark raw outputs, model/checkpoint artifacts, biometric/capture data, creative/canon drafts, service configuration, raw test/counterexample bodies, dirty patch/reject variants, cache/build/vendor directories, and too-skeletal placeholders.
  • Updated index.md News and Queries sections for the new nightly desk page; content-page count moved 174 → 175. Pre-existing uncommitted 2026-05-16 nightly desk files and index/log lines were in the same news-desk family and were retained for a coherent index; unrelated East Bay property-monitor working-tree changes and wiki .hermes/ material remained untouched and uncommitted.
  • Lint result: bash scripts/lint-wiki.sh passed after the new page, index, and log edits: 175 content pages checked.

[2026-05-27] query | nightly src projects desk

  • Created raw survey note raw note and query page nightly-src-projects-desk-2026-05-27 for the nightly local-projects desk.
  • Published verdict: gemma-dungeon, jepa-lang, and jepa-poker are tonight’s clearest technical leads; testing-rl remains the sturdy verifier bench; deer-flow, openai-symphony, Gas City/Basis/Steward, parenting-bookshelf-compass, handterm, and selected side rooms were summarized only where inspectable evidence supported public-safe claims.
  • Generated editorial illustration queries/news-assets/2026-05-27-project-desk-hero.svg as deterministic symbolic SVG after rejecting raster drafts with text artifacts; no fake screenshots were used, and the SVG parsed as valid XML.
  • Survey automation status: exactly 10 top-level Hermes survey lane identities ran in one batch across all 50 top-level directories under the local src tree, including hidden directories. All 10 lanes reported three read-only subteams for purpose/docs/manifests, live-work evidence, and public-safety review, plus one further three-way leaf probe. A controller audit found 50 assigned directories, 50 unique assignments, no missing directories, no extras, and no duplicates.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local settings, security/dependency scan artifacts, empty or skeletal directories, provocative/protected-class-sensitive social-claim material, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator/oracle payloads, raw benchmark outputs, model/checkpoint artifacts, biometric/capture data, creative/canon/world-packet drafts, service configuration, raw test/counterexample bodies, dirty patch/reject variants, cache/build/vendor directories, and too-skeletal placeholders.
  • Updated index.md News and Queries sections for the new nightly desk page; committed content-page count moved 175 → 176. Pre-existing uncommitted source-essence, East Bay property-monitor, and wiki .hermes/ working-tree material remained untouched and uncommitted.
  • Lint result: bash scripts/lint-wiki.sh passed in the working tree before staging: 213 content pages checked; committed-tree verification passed after commit: 176 content pages checked.

[2026-05-29] query | nightly src projects desk

  • Created raw survey note raw note and query page nightly-src-projects-desk-2026-05-29 for the nightly local-projects desk.
  • Published verdict: gas-city-but-its-just-codex is tonight’s clearest orchestration/control-plane lead; cardgame1 and gemma-dungeon carry the game/research craft story; testing-rl remains the verifier bench; kettlebellsim is the clean simulation-validation lead; unconventional-jepa-lab, jepa-lang, textual-world-model, selected NNPL projects, parenting-bookshelf-compass, and handterm were summarized only where inspectable evidence supported public-safe claims.
  • Generated editorial illustration queries/news-assets/2026-05-29-project-desk-hero.svg as deterministic symbolic SVG after rejecting two generated raster drafts with text artifacts; no fake screenshots were used, and the SVG parsed as valid XML.
  • Survey automation status: exactly 10 top-level Hermes survey lane identities ran in one batch across all 50 top-level directories under the local src tree, including hidden directories. All 10 lanes reported three read-only subteams for purpose/docs/manifests, live-work evidence, and public-safety review, plus one further three-way leaf probe. A controller audit found 50 assigned directories, 50 unique assignments, no missing directories, no extras, and no duplicates.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local assistant/settings directories, security/dependency scan artifacts, empty or skeletal directories, provocative/protected-class-sensitive social-claim material, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator/oracle payloads, raw benchmark outputs, model/checkpoint artifacts, biometric/capture data, creative/canon/world-packet drafts, service configuration, raw test/counterexample bodies, dirty patch/reject variants, cache/build/vendor directories, local .env-style material, and too-skeletal placeholders.
  • Updated index News and Queries sections for the new nightly desk page. Working-tree content-page count moved 213 → 214; the isolated committed-tree update moves 176 → 177 because pre-existing uncommitted source-essence, East Bay property-monitor, and wiki .hermes/ material remained untouched and uncommitted.
  • Lint result: bash scripts/lint-wiki.sh passed in the working tree before staging: 214 content pages checked; committed-tree verification passed after commit: 177 content pages checked.

[2026-05-31] query | nightly src projects desk

  • Created raw survey note raw note and query page nightly-src-projects-desk-2026-05-31 for the nightly local-projects desk.
  • Published verdict: gemma-dungeon is tonight’s clean same-night lead; cardgame1, testing-rl, testing-rl-hermes, and kettlebellsim are the strongest public-safe working set; Basis/Hermes, another-harness, is-it-formal, openai-symphony, steward, JEPA/NNPL/tinygrad research benches, handterm, and parenting-bookshelf-compass were summarized only where inspectable evidence supported public-safe claims.
  • Generated editorial illustration queries/news-assets/2026-05-31-project-desk-hero.svg as deterministic symbolic SVG after the image backend reported missing FAL_KEY; no fake screenshots were used, and the SVG parsed as valid XML.
  • Survey automation status: exactly 10 top-level Hermes survey lane identities ran in one batch across all 50 top-level directories under the local src tree, including hidden directories. All 10 lanes reported three read-only subteams for purpose/docs/manifests, live-work evidence, and public-safety review, plus one further three-way leaf probe. A controller audit found 50 assigned directories, 50 unique assignments, no missing directories, no extras, and no duplicates.
  • Public-safety filter: project-specific public detail was held back or reduced to category-only mention for hidden local assistant/settings directories, security/dependency scan artifacts, empty or skeletal directories, provocative/protected-class-sensitive social-claim material, local deployment/model-runner folders, private corpus bodies, prompt/agent/skill instruction bodies, scratch/meta workspaces, generated media, raw logs/prompts/trajectories, evaluator/oracle payloads, raw benchmark outputs, model/checkpoint artifacts, biometric/capture data, creative/canon/world-packet drafts, service configuration, raw test/counterexample bodies, dirty patch/reject variants, cache/build/vendor directories, local .env-style material, and too-skeletal placeholders.
  • Updated index News and Queries sections for the new nightly desk page. Working-tree content-page count moved 214 → 215; the isolated committed-tree update moves 177 → 178 because pre-existing uncommitted source-essence, East Bay property-monitor, and wiki .hermes/ material remained untouched and uncommitted.
  • Lint result: bash scripts/lint-wiki.sh passed in the working tree before staging: 215 content pages checked; committed-tree verification passed after staging: 178 content pages checked.

Graph View

Backlinks