feat: expand internal skills across agent harnesses

[sparkus]

Intent

Make firstmate's internal skills discoverable on grok and codex, not just claude, and add one new agent-only skill for ticket-queue discipline. This is firstmate's own shared tracked material.

VERIFY first (do not skip): confirm with evidence what grok, codex, and claude actually resolve for project-level skills - which paths each harness reads, that a symlink is not assumed to work without observation, and that git preserves mode-120000 symlinks. Record findings in the PR body / maintainer verification doc. Key observed fact: a grok 0.2.118 session in this worktree loaded firstmate-coding-guidelines from the project path (.../.grok/skills/... which is a symlink to .agents/skills), NOT from ~/.grok/skills (skill absent from user home); confirmed via grok inspect --json source.type=project. Codex codex-cli 0.146.0-alpha.9.2 via codex debug prompt-input listed skill root r0=.../firstmate/.agents/skills and Available skills including firstmate-coding-guidelines and ticket-queue-discipline. Claude discovery remains via .claude/skills symlink. All three harness skill paths committed as git mode 120000.

Wire discovery so .agents/skills remains the single canonical source (no duplicated skill content). Match the claude pattern for .codex/skills and .grok/skills symlinks to ../.agents/skills. Document verified evidence.

Add ONE agent-only internal skill ticket-queue-discipline under .agents/skills with metadata.internal: true and user-invocable: false. Trigger: before claiming, dispatching, closing, or auditing tracker-backed ticket work. Content is JUDGMENT ONLY: one authoritative live tracker; claim atomically at dispatch; verify acceptance criteria against code and merge history at pickup; close with satisfied-by naming the merged PR; decision classes with one-line reasons; convergence; every decision becomes an executable test; evidence honesty.

Acceptance: verified evidence of what each harness resolves in PR body; single canonical .agents/skills; new skill loads on all three harnesses; existing claude discovery unbroken and proven by CI symlink invariants. Stay inside skill discovery plus the one new skill.

PR must open against captain fork (sparkus/firstmate), base origin/main - not upstream kunchenguid.

What Changed

• Add Codex and Grok project-discovery symlinks to the canonical .agents/skills tree, with CI invariants matching Claude so all three harnesses share one tracked source.
• Add the agent-only ticket-queue-discipline skill for live-tracker claims, acceptance and merge verification, decision escalation, convergence, executable tests, and honest close-out evidence.
• Document verified runtime discovery for Claude, Codex, and Grok, including resolved project paths, symlink behavior, and git mode 120000 preservation.

Risk Assessment

✅ Low: Captain, the prior finding is resolved with honest Claude runtime evidence, the explicitly authorized post-merge confirmation remains clearly bounded, and the source wiring is canonical and well-contained.

Testing

Baseline scope checks, live three-harness discovery, project-versus-user provenance, symlink and canonical-source verification, and the focused documentation contract passed; initial budget-capped Claude probes were retried successfully without the artificial cap, reviewer-visible evidence was captured, and the raw debug log was replaced by a sanitized excerpt.

Evidence: Live Claude/Codex/Grok discovery transcript

^D��Harness versions:
grok 0.2.118 (1e1687c1cf6a) [stable]
codex-cli 0.146.0-alpha.9.2
2.1.220 (Claude Code)
Grok project skill resolution:
{
  "skill": {
    "name": "ticket-queue-discipline",
    "description": "Agent-only judgment for tracker-backed ticket work. Load before claiming, dispatching, closing, or auditing tickets in a project tracker so readiness, claims, and close reasons stay tied to live canon rather than scratch state.",
    "source": {
      "type": "project",
      "path": "/Users/agwerschky/.no-mistakes/worktrees/e661563b090a/01KZ4B0458NGJFN2BJ5QTM3VFN/.grok/skills/ticket-queue-discipline/SKILL.md"
    },
    "userInvocable": false
  }
}
Codex project skill resolution:
- `r0` = `/Users/agwerschky/.no-mistakes/worktrees/e661563b090a/01KZ4B0458NGJFN2BJ5QTM3VFN/.agents/skills`
- ticket-queue-discipline: Agent-only judgment for tracker-backed ticket work. Load before claiming, dispatching, closing, or auditing tickets in a project tracker so readiness, claims, and close reasons sta (file: r0/ticket-queue-discipline/SKILL.md)
Claude project skill invocation:
{"type":"assistant","content":[]}
{"type":"assistant","content":[{"type":"tool_use","id":"toolu_01Fxx6vSWX3kyjSYevRNjZ3F","name":"Skill","input":{"skill":"ticket-queue-discipline"},"caller":{"type":"direct"}}]}
{"type":"user","content":[{"type":"tool_result","tool_use_id":"toolu_01Fxx6vSWX3kyjSYevRNjZ3F","content":"Launching skill: ticket-queue-discipline","is_error":null}]}
{"type":"user","content":[]}
{"type":"assistant","content":[{"type":"text","text":"LOADED ticket-queue-discipline."}]}
{"type":"result","subtype":"success","result":"LOADED ticket-queue-discipline.","is_error":false}
�[?25h

Evidence: Mode-120000 and canonical-source transcript

^D��Target commit: 5939a39fbf2e52a6fe8e22a56d42cfdc2c8a65b0
Tracked harness link modes and blob identity:
120000 2b7a412b8fa0fb7e985b0793321bd4e698f2b6cd 0	.claude/skills
120000 2b7a412b8fa0fb7e985b0793321bd4e698f2b6cd 0	.codex/skills
120000 2b7a412b8fa0fb7e985b0793321bd4e698f2b6cd 0	.grok/skills
Resolved link targets:
.claude/skills -> ../.agents/skills
.codex/skills -> ../.agents/skills
.grok/skills -> ../.agents/skills
Tracked ticket skill sources:
.agents/skills/ticket-queue-discipline/SKILL.md
Frontmatter contract:
---
name: ticket-queue-discipline
description: >-
  Agent-only judgment for tracker-backed ticket work.
  Load before claiming, dispatching, closing, or auditing tickets in a project tracker
  so readiness, claims, and close reasons stay tied to live canon rather than scratch state.
user-invocable: false
metadata:
  internal: true
---

# ticket-queue-discipline

Evidence: Grok project provenance transcript

^D��/Users/agwerschky/.grok/skills/firstmate-coding-guidelines/SKILL.md absent
/Users/agwerschky/.grok/skills/ticket-queue-discipline/SKILL.md absent
{
  "skill": {
    "name": "ticket-queue-discipline",
    "source": {
      "type": "project",
      "path": "/Users/agwerschky/.no-mistakes/worktrees/e661563b090a/01KZ4B0458NGJFN2BJ5QTM3VFN/.grok/skills/ticket-queue-discipline/SKILL.md"
    },
    "userInvocable": false
  }
}

Evidence: Sanitized Claude project-root and SkillTool excerpt

^D��24:2026-08-03T17:47:16.949Z [DEBUG] Loading skills from: managed=/Library/Application Support/ClaudeCode/.claude/skills, user=/Users/agwerschky/.claude/skills, project=[/Users/agwerschky/.no-mistakes/worktrees/e661563b090a/01KZ4B0458NGJFN2BJ5QTM3VFN/.claude/skills]
229:2026-08-03T17:47:22.578Z [DEBUG] SkillTool returning 2 newMessages for skill ticket-queue-discipline
230:2026-08-03T17:47:22.578Z [INFO] [Stall] tool_dispatch_end tool=Skill toolUseId=toolu_01YGMQ7hkxPCC5TUbaFokkFm outcome=ok durationMs=1

Evidence: Focused documentation contract test

^D��FM_TEST_BEGIN 2026-08-03T17:51:52Z tests/fm-documentation-audiences.test.sh family=pure-contract-unit expected_gate_skip=none
ok - documentation inventory classifies every maintained prose surface exactly once
ok - classification, setup routing, and maintained-prose scope fail safely
ok - required documentation owner pointers cannot silently disappear
ok - local links resolve while dates, versions, commands, and incident prose remain semantically reviewed
FM_TEST_END 2026-08-03T17:51:53Z tests/fm-documentation-audiences.test.sh exit=0 duration_ms=624 gate_skip=false
FM_TEST_SUMMARY total=1 failed=0 skipped_gate=0 duration_ms=676
FM_TEST_SUMMARY_FAMILY family=pure-contract-unit count=1 duration_ms=624 failed=0
FM_TEST_SLOWEST rank=1 script=tests/fm-documentation-audiences.test.sh duration_ms=624

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

🔧 **Review** - 1 issue found → auto-fixed ✅

• 🚨 docs/verification/harness-skill-discovery.md:71 - Intent requires “confirm with evidence what ... claude actually resolve[s],” says a symlink must not be assumed to work without observation, and requires the new skill to load on all three harnesses. This Claude section only runs readlink, checks the pre-existing afk file, and inspects Git mode; it never observes Claude resolving or loading ticket-queue-discipline. Record Claude runtime discovery evidence for the new project skill, or obtain explicit approval to treat filesystem/CI invariants as the proxy.

🔧 Fix: Record Claude runtime skill resolution evidence
✅ Re-checked - no issues remain.

✅ **Test** - passed

✅ No issues found.

• git status --short --branch; git rev-parse HEAD; git diff --name-status 4e01d7fc449d4ffff932f172721f1fea53dfa6e6..5939a39fbf2e52a6fe8e22a56d42cfdc2c8a65b0
• grok inspect --json filtered to ticket-queue-discipline, with user-home absence checks
• codex debug prompt-input "x" filtered to project root r0 and ticket-queue-discipline
• claude -p --output-format stream-json --verbose --no-session-persistence --permission-mode dontAsk --tools Skill ...
• git ls-files -s .claude/skills .codex/skills .grok/skills with readlink, readability, canonical-body cmp, and tracked-source checks
• bin/fm-test-run.sh tests/fm-documentation-audiences.test.sh after the documentation correction

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.