chore: add Codex discovery for PyAutoBrain agents

[Jammy2211]

Overview

Make every public PyAutoBrain conductor and faculty discoverable as a Codex skill while preserving the existing Claude command surface. Extend the organism-wide installer to support both harnesses without duplicating canonical workflow bodies.

This is phase 1 of the approved Codex-wrapper rollout. Phase 2 applies the resulting contract to PyAutoMind, PyAutoHeart, and PyAutoBuild.

Plan

• Add thin SKILL.md wrappers for every public PyAutoBrain agent and existing command/workflow surface.
• Update the installer to install skills for both Claude and Codex while retaining Claude commands when both files exist.
• Remove shared Brain wording that assumes Claude-only models, tools, or generated output.
• Add isolated validation for metadata, public-agent coverage, symlink installation, and existing behavior.
• Run the autonomous ship gate and stop at an open PR.

Detailed implementation plan

Affected Repositories

• PyAutoBrain (primary)

Branch Survey

Repository	Current Branch	Dirty?
PyAutoBrain	main	clean

No active worktree claims conflict with PyAutoBrain.

Suggested branch: feature/codex-brain-skill-wrappers

Work classification: organism infrastructure (direct single-repo maintenance)

Worktree root: /home/jammy/Code/PyAutoLabs/.worktrees/codex-brain-skill-wrappers/

Implementation Steps

1. Add concise wrappers beside existing command bodies under skills/*/SKILL.md, pointing to the canonical Markdown workflow and CLI rather than copying it.
2. Add skill directories for public CLI agents without an existing command surface: clone, release, vitals, review, memory, and samplers.
3. Update bin/install.sh so skill and command discovery are independent, Claude receives both surfaces where present, and Codex receives every skill.
4. Parameterize installation roots so tests can use temporary homes without touching live ~/.claude or ~/.codex directories.
5. Update AGENTS.md, bin/README.md, skills/COMMANDS.md, and shared workflow references to describe both harnesses and use model-tier terminology.
6. Add pytest coverage for dispatcher-to-skill completeness and dual-harness installation.
7. Validate every wrapper with Codex quick_validate.py, run the skill line-count guard, shell syntax checks, and the PyAutoBrain test suite.

Key Files

• bin/pyauto-brain - authoritative public agent inventory.
• bin/install.sh - cross-organ skill and command installer.
• skills/*/SKILL.md - thin Codex discovery wrappers.
• skills/COMMANDS.md - command/agent surface contract.
• skills/WORKFLOW.md - cross-harness workflow doctrine.
• tests/ - isolated installer and coverage tests.

Original Prompt

Click to expand starting prompt

yes, make SKILL.md wrappers for PyAutoBrain and any other repos you see a Claude bias

Autonomy

Launched with --auto; effective level: safe (maintenance cap). The run may proceed through the four-leg autonomous ship gate and ends at PR-open. Merge and issue closure remain human actions.

[Jammy2211]

Autonomous ship gate parked at Heart acknowledgement

Implementation and branch-local validation are complete:

• Tests: 30 passed
• Smoke: n/a — organism skill/installer changes have no downstream script surface
• Independent review: CLEAN after resolving one installer compatibility finding
• Heart: YELLOW, not acknowledged for this launch

To permit the branch push and PR creation, please explicitly acknowledge this exact current Heart reason set:

1. workspace validation not passing (3 failed, 2026-07-09)
2. 58 stale parked scripts
3. autolens_assistant pinned behind the installed 2026.7.6.649 stack
4. tenant-firewall manifest drift (5 mismatches against PyAutoMind/repos.yaml)
5. PyAutoMind has an open PR 11 days old

Recommendation: acknowledge these organism-level reasons for this task launch only. Branch-specific correctness is covered by tests and the CLEAN review. Any new Heart reason or RED verdict will still stop the run.

Without acknowledgement, commit 619457b remains local and nothing is pushed.

[Jammy2211]

Phase 1 remains locally complete: 30 tests passed, all skill metadata and line-budget checks passed, and independent review is CLEAN. No source/API changes; commits 2150afa and 619457b remain local and unpushed.\n\nCurrent Heart gate (2026-07-11): YELLOW, score 42.\n\nExact reasons:\n- workspace validation not passing (3 failed, 2026-07-09T09-48-30Z)\n- 58 stale parked script(s)\n- autolens_assistant: pinned BEHIND installed 2026.7.6.649\n- manifest drift: tenant firewall (organ code) - 5 mismatch(es) vs PyAutoMind/repos.yaml\n- PyAutoMind: open PR 11d old\n- install verification not run\n- no release validation for current source\n\nUnder AUTONOMY.md, the branch remains local and unpushed until these exact YELLOW reasons are explicitly acknowledged.

[Jammy2211]

Phase 1 remains locally complete: 30 tests passed, skill metadata and line-budget checks passed, independent review CLEAN; commits 2150afa and 619457b remain local and unpushed.

Current Heart gate (2026-07-11): YELLOW, score 42.

Exact current reasons:

• workspace validation not passing (3 failed, 2026-07-09T09-48-30Z)
• 58 stale parked script(s)
• autolens_assistant: pinned BEHIND installed 2026.7.6.649
• PyAutoMind: open PR 11d old
• install verification not run
• no release validation for current source

The prior seven-reason checkpoint is superseded because the tenant-firewall manifest drift has cleared. Under AUTONOMY.md, the branch remains local and unpushed until this exact six-reason YELLOW set is explicitly acknowledged.