Agent Harness Wiki

8 min read

Nightly Src Projects Desk Raw Survey (2026-05-10)

This raw note preserves the public-safe basis for the 2026-05-10 nightly src/ projects desk. It summarizes inspectable repository evidence only: README/docs, manifests, branch and commit metadata, status summaries, safe filenames, mtimes, tests, plans, and visible checked-in artifacts. It does not publish secret-bearing files, local settings, raw prompts/logs/trajectories, private corpus bodies, evaluator payloads, raw benchmark outputs, checkpoint/model artifacts, biometric/capture data, generated media bodies, or sensitive/provocative material.

Where a directory itself is local, sensitive, private-corpus-backed, or too skeletal, this note uses category-level wording. The desk is a survey, not an excuse to turn a source tree into confetti.

Survey scope and method

  • Survey root: /Users/ericfode/src.
  • Survey timestamp: 2026-05-10.
  • Full top-level directory count: 38, including hidden directories.
  • Execution shape: exactly 10 top-level Hermes survey lane identities, dispatched as one batch of 10.
  • Lane recursion: all 10 lane summaries reported delegate_task availability. Each lane spawned a three-way survey team for purpose/docs/manifests, live-work evidence, and public-safety eligibility, and each of those subteams reported one further three-way leaf recursion. Additional recursion ended at the configured depth cap.
  • Controller audit: a post-dispatch read-only audit corrected an inventory spelling slip for the three nnpl-* directories. The top-level lane count remained exactly 10; the corrected audit covered nnpl-external-latent-bus, nnpl-shared-bus, and nnpl-typed-boundary-ir using the same evidence rules.
  • Evidence allowed: README/docs/plans, manifests, branch/status/log metadata, safe modified/untracked filenames, mtimes, tests, checked-in reports, and visible artifacts.
  • Evidence excluded: secret contents, .env contents, hidden local settings, raw prompts/logs/trajectories, hidden evaluator/supervisor payloads, private corpus bodies, explicit/provocative/unsafe material, raw benchmark outputs, checkpoints/model artifacts, biometric/capture data, generated media bodies, and directories too skeletal for a responsible public claim.
  • Illustration: generated locally as symbolic SVG editorial art at queries/news-assets/2026-05-10-project-desk-hero.svg; it is not a screenshot.

Ten survey lanes

  1. Hidden local settings; hidden tinygrad research checkout; another-harness; one sensitive social-claim wiki withheld by category.
  2. basis; basis-hermes; basis-jcode; cardgame1 / Dungeon Steward.
  3. Empty creative; deer-flow; FACEMUSIC; gas-city-but-its-just-codex.
  4. gemma4-tinygrad-opt; handterm; hoid; is-codex-better.
  5. is-it-formal; justfooln; kettlebellsim; hidden-only/skeletal Kimi settings directory.
  6. Local langfuse; local-hermes; scratch meta-hermes; corrected audit for nnpl-external-latent-bus.
  7. Corrected audit for nnpl-shared-bus; corrected audit for nnpl-typed-boundary-ir; openai-symphony; empty overengineeredlife.
  8. silly-pi-stuff; private spec-dataset-evolution-corpus; nested src skill scaffold; steward.
  9. testing-rl; testing-rl-hermes; empty tinygrad.
  10. tinygrad-gemma; empty tinygrad-gemma-gemini; tinygrad-gemma-kimi.

Public-safe lead candidates

Harness, orchestration, and formal surfaces

  • another-harness evidence: git repo on main with no commits yet and hundreds of untracked entries. Safe files include README.md, Lean/Lake metadata, AnotherHarness.lean, docs, tests, tools, benchmarks, and plugins. The README describes a Codex-native harness / Hermes-competition proving ground while preserving a Lean scaffold. Safe summary: an early all-untracked harness/formalization workspace; state, handoffs, evaluations, prompts, trajectories, generated media, raw artifacts, and temp/cache trees are withheld.
  • gas-city-but-its-just-codex evidence: dirty Rust/Lean/operator repo on codex/native-codex-ui, HEAD 198aefc from 2026-04-21. Key evidence includes README, Rust workspace, workflow-ledger specifications, schemas/templates, MCP/gRPC/app-server surfaces, operator tooling, docs/scripts, tests, and formal Lean material. Safe summary: Codex-native durable workflow/control-plane research prototype. Runtime state, logs, transcripts, database files, context boards, benchmark payloads, workflow/thread IDs, and live operator state are withheld.
  • openai-symphony evidence: dirty Elixir/Phoenix repo on main, HEAD 58cf97d from 2026-04-27, with README, SPEC, license, Elixir docs, Mix manifest, LiveView/API/dashboard/logging/token-accounting material, and tests. Safe summary: engineering-preview orchestration service for turning issue-tracker work into isolated autonomous implementation runs. Logs, prompt/workflow bodies, hidden tooling, local runtime details, and unreviewed diffs are withheld.

Specification, Basis, and spec-code grounding

  • basis evidence: clean Elixir/Mix repo on main, HEAD a5544e0 from 2026-05-07. Safe files include spec.md, Mix metadata, reducer and implementation-imaginer component specs, docs, and tests. Safe summary: project for reducing prose/spec artifacts into structured, provenance-backed specification state.
  • basis-hermes evidence: clean Python/Hermes plugin repo on main, HEAD 0061d32 from 2026-05-05, with README.md, plugin.yaml, pyproject.toml, dashboard manifest, reducer/validator source, CLI/tool handlers, and tests. Safe summary: Hermes-native wrapper exposing deterministic Basis reducer and packet-validator surfaces.
  • basis-jcode evidence: git repo on main, HEAD 4b1e621 from 2026-05-05, ahead of origin and dirty with tracked deletions in reducer example/UI files. Safe summary: Jcode-native reducer/control-plane variant for ledgers, validation, UI projections, and dashboard decision flows. Raw .basis run trees, prompts, event streams, worker packets, validation bodies, and run graphs are withheld.
  • steward evidence: clean design-stage repo on main, HEAD ba88837 from 2026-05-05, with README, Python manifest, docs for charter, architecture, benchmark spec, implementation plan, data governance, roadmap, workflows, and decision log. Safe summary: local-first spec-code grounding and benchmark design project; private-corpus-derived bodies and example packets remain withheld.
  • is-it-formal evidence: unborn git repo with Lean/Lake scaffold, README, lakefile.toml, lean-toolchain, examples, and Python CLI tooling. Safe summary: early Lean-backed scaffold for grading how formal a claim really is.

Test-writing and verifier environments

  • testing-rl evidence: git repo on master, HEAD 139cea4 from 2026-05-04, dirty with modified workflow/docs/scripts and untracked recent-data page/test material. Key evidence includes README, SPEC, pyproject, docs dashboard, environment contract, artifact schemas, risk/replay/counterfactual docs, Hermes/Atropos/Tinker adapter docs, Lean files, benchmark task filenames, and tests. Safe summary: prototype RL/test-generation environment for writing high-value software tests against hidden reference behavior. It explicitly warns local candidate-test execution is not a security sandbox.
  • testing-rl-hermes evidence: clean local repo on main with no remote configured, recent commits from 2026-05-01/02, MASTER_PLAN.md, adversarial risk review, test-generation environment docs, history-derived fixture docs, reports, source, and tests. Safe summary: Hermes-oriented/history-derived test-generation RL prototype; not presented as public, licensed, or fully validated.
  • Hidden references, oracle/mutant bodies, evaluator internals, raw benchmark JSON, prompt trajectories, and replay artifact bodies remain withheld.

Model, tinygrad, and NNPL benches

  • tinygrad-gemma evidence: git repo on main, ahead of origin/main by 93 commits, no tracked source modifications, many untracked benchmark/local artifact entries, README, pyproject, docs/configs/benchmarks/scripts/tests, and console scripts for CLI/chat. Recent commits from 2026-05-06/07 record worker rounds. Safe summary: native tinygrad package/runtime for Gemma 4 checkpoints, text and multimodal preprocessing, tokenizer support, generation, CLI/chat interfaces, Metal support, and checkpoint/quantization/fine-tuning utilities. Raw checkpoints, benchmark logs, and performance claims are withheld.
  • gemma4-tinygrad-opt evidence: non-git optimization sandbox with nested clean tinygrad checkout at e9983e3, local tinygrad_gemma/ package, model/loader/tokenizer scripts, and benchmark/kernel scripts. Safe summary: local Gemma/tinygrad optimization and inference sandbox; root lacks a README and artifact bodies remain private.
  • tinygrad-gemma-kimi evidence: dirty git repo on opt/attention, no remote, commits from 2026-04-25/26, modified attention/validation/benchmark scripts, generated caches, patches, results, and no README/manifests. Safe summary: category-level tinygrad/Gemma attention/MoE/JIT/correctness scratch work only.
  • nnpl-external-latent-bus evidence: non-git Python/Numpy prototype with README, pyproject, docs, source, tests, project brief, and artifacts. Safe summary: two-space external/internal latent bus architecture, with explicit bridges and option-preserving planning benchmarks.
  • nnpl-shared-bus evidence: non-git scaffold with README, docs, configs, source, tests, and run directories. README records an honest negative or insufficient v0 result for the shared-bus thesis and an automated test-suite receipt. Safe summary: one-bus NNPL variant with strong baselines and falsification-oriented run artifacts; raw run metrics/traces/readouts are withheld.
  • nnpl-typed-boundary-ir evidence: non-git Python/tinygrad prototype with README, pyproject, docs, data/results/scripts/source/tests, and typed IR spec material. Safe summary: typed-boundary NNPL variant for legality, auditability, deterministic rendering, and failure localization. Raw result bodies and exact benchmark examples are withheld.

Simulation, terminal, interface, and craft work

  • kettlebellsim evidence: clean git repo on codex/reward-audit-and-swing-training, recent commits on 2026-05-09 around Modal/Isaac/planar handoff work, package manifest, docs, scripts, configs, recipes, skills, and extensive tests. Safe summary: simulation-first kettlebell path-signature and biomechanics research toolkit, with remote simulator/RL scaffolding. Trajectories, media, reports, service config, and prompt-like council material are withheld.
  • handterm evidence: clean Rust repo on master, HEAD 977e709 from 2026-04-19, README, Cargo workspace, optimization/remain-work docs, tests, and recent graphics/kitty-upload refactors. Safe summary: Wayland-native terminal emulator focused on performance, renderer paths, and low-overhead multi-window architecture.
  • FACEMUSIC evidence: dirty Rust/web/iOS/ML repo, HEAD f6cf6cf from 2026-04-19, with browser architecture docs, web package manifest, iOS README, ML README, Rust tests, and modified iOS/web/audio/control files. Safe summary: category-only privacy-sensitive face-controlled music/instrument prototype. Biometric captures, recordings, model checkpoints, saliency/probe outputs, sessions, and generated media are withheld.
  • cardgame1 / Dungeon Steward evidence: clean Godot 4.6 repo on hermes/combat-stage-art-fallback-upstream, HEAD a9a8ef6 from 2026-04-15, with game design docs, Godot project metadata, source/data folders, deterministic/simulation/smoke tests, and generated-art workflow paths. This run keeps it out of project-specific public detail because ignored env/session-log paths, agent scaffolding, prompt/image-generation paths, and generated artifact paths dominate the safety surface.
  • hoid evidence: dirty private creative/tooling repo with Go/game/Lean/test surfaces and modified story/world files. Safe treatment is category-only; story/canon/comic/music bodies and generated media assets are withheld.

Held back from project-specific public detail

The survey fully held back, or reduced to category-only mention, hidden local settings, hidden-only or empty directories, one sensitive social-claim wiki, local deployment/model-runner folders, a private spec corpus, 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, story/canon drafts, local service configuration, and cache/build/vendor directories.

Editorial synthesis

The publishable movement clusters around five themes:

  1. harness/control-plane work is increasingly repo-artifact-first, with ledgers, dashboards, app-server bridges, and Lean/formal surfaces rather than only chat rituals;
  2. specification work is moving toward reducer packets, provenance, and deliberately smaller conceptual state;
  3. test-generation environments are making reward, replay, hidden-reference, and sandbox limits visible;
  4. tinygrad/Gemma and NNPL benches are active but remain behind artifact and benchmark-safety gates;
  5. simulation, terminal, and interface projects keep enough docs/tests/manifests to be described without mysticism, while private creative and biometric material stays behind the curtain.

A public note can say that much. It should not say more merely because the filesystem is willing to be found.

Graph View

Backlinks