feat(hygiene): hygiene conductor — phase 1 scaffold + boundaries (#87)

[Jammy2211]

Closes #87 (phase 1 of the hygiene conductor).

What

Phase 1 of the PyAutoBrain hygiene conductor — the code-quality upkeep
function (developer-loop cost + repo tidiness), distinct from Heart
(proof-of-works) and profiling (modelling speed). This makes the conductor
real, routable, and correctly bounded, with its five modes staged as stubs —
no behaviour yet. Design of record: PyAutoMind/feature/pyautobrain/hygiene_agent.md;
decision (conductor, no new paired repo): PyAutoMind/research/pyautobrain/hygiene_agent_decision.md.

Changes

• agents/conductors/hygiene/{AGENTS.md, hygiene.sh} — the conductor + a
  stdlib/bash skeleton emitting a staged HygieneDecision (default worklist +
  perf/tidy/noise/deps/docs; --json footing).
• bin/pyauto-brain — dispatch map, description, usage comment, CONDUCTOR_ORDER.
• skills/hygiene/{SKILL.md, hygiene.md} — the /hygiene veneer + Codex wrapper.
• skills/COMMANDS.md — /hygiene veneer entry + tier-1 conductor row.
• agents/conductors/profiling/AGENTS.md — rewrite the vs hygiene boundary to
  the measured-thing split (profiling = modelling speed; hygiene = dev-loop cost)
  and move "hunting generally-slow functions flagged by integration tests"
  into hygiene's perf mode.
• skills/repo_cleanup/SKILL.md — repoint the "Future Cleanup Agent" note at the
  built conductor (absorbed as tidy in phase 2).

Out of scope (later phases): phase 2 absorbs repo_cleanup (tidy) +
cli_noise_clean (noise) and consults dep_audit (deps) / audit_docs
(docs); phase 3 builds perf (test/script/import timing) + optional PyAutoHeart
observation legs (import_time / cli_noise).

Autonomous-ship gate (--auto, effective level: supervised)

• Tests — PyAutoBrain suite 30 passed. Downstream library suites: n/a
  (Brain is the reasoning/org repo; no Python public API consumed by the science
  libraries — no python-source review flag).
• Smoke — n/a: Brain has no downstream workspace-script surface
  (smoke_tests.txt covers the six science workspaces).
• Review — reviewing-agent verdict CLEAN (wiring + docs + a bash stub;
  the one nit — hardcoded help line-numbers — was fixed to parse the # Usage:
  block). The no-test-changes risk flag is covered by test_skill_install,
  which now asserts hygiene is a well-formed public agent.
• Heart — organism YELLOW (PyAutoBrain behind, unrelated dirty workspaces);
  no reason set acknowledged at launch. Not GREEN — this run parked at ship
  sign-off (supervised) and shipped on explicit human go-ahead, not an
  autonomous safe-ship.

Validation checklist (reviewer)

• bin/pyauto-brain hygiene prints the staged 5-mode menu; a mode prints its staged notice.
• bin/pyauto-brain hygiene --json is valid JSON; hygiene <mode> --json likewise.
• bin/pyauto-brain help lists hygiene in the conductor block; help hygiene renders the charter.
• /hygiene resolves after bash bin/install.sh (auto-discovers skills/hygiene/).
• profiling's vs hygiene boundary + Future-modes read correctly; repo_cleanup pointer updated.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Tw3EwV55k6VzxorYng3Kfn