[codex] prepare public release harness docs

README.md

@@ -263,6 +263,9 @@ Dotagents keeps `~/.agents` as the source of truth and adapts each agent through
 | Hermes | Config path to `~/.agents/skills` | Not managed | `~/.hermes/config.yaml` | `~/.hermes/config.yaml` for known lifecycle hooks | Reads configured Hermes context | Uses `skills.external_dirs`; do not mirror into `~/.hermes/skills` because bundled categories can collide. |
 | Factory Droid | Symlink mirror to `~/.factory/skills` | Generated to `~/.factory/droids` | `~/.factory/mcp.json` | `~/.factory/settings.json` | `~/.factory/AGENTS.md` symlink | Full managed mirror for skills, roles, MCP, and supported hooks. |
 | Pi/OMP | Symlink mirror to `~/.omp/agent/skills` | Not managed | `~/.omp/agent/mcp.json` | Not managed | Reads configured OMP context | Primary OMP target for shared skills, MCP entries, and portable plugin skill surfaces. |
+| Amp | Compatibility-only via explicit local config | Not managed | Supported by CLI when locally targeted | Not managed | Not managed | Intentionally absent from canonical `dotagents.yaml`; use a gitignored local overlay for one-off migration or preservation work. |
+| OpenCode | Not managed | Not managed | Not managed | Not managed | Not managed | Compatibility research only; no verified native skill/MCP surface is managed by dotagents yet. |
+| OpenClaw | Not managed | Not managed | Not managed | Not managed | Not managed | Compatibility research only; needs an owner and verified config surface before dotagents treats it as managed. |
 
 Compatibility-only harness support may remain in the CLI for migration, hook cleanup, trailer stripping, or one-off local configs. Those harnesses are intentionally absent from the canonical `dotagents.yaml` managed target list.
 

SPEC.md

@@ -0,0 +1,121 @@
+# SPEC: dotagents public release
+
+Executor: `codex exec` (GPT-5.5-high), one milestone per run, in order. Each milestone has machine-checkable acceptance criteria; do not start the next milestone until the current one passes. Human gates are marked HUMAN GATE and must stop the run with a clear summary for the user.
+
+## Goal
+
+Make this repo public and presentable: a stranger can (a) install the portable skills into Claude Code as a plugin, (b) install the `dotagents` CLI and run their own canonical `~/.agents` repo using this one as the living reference, and (c) read a single GitHub Pages landing page that shows the project and three happy-path scenarios.
+
+Positioning: tool + living example. The landing sells the `dotagents` CLI and the canonical-repo pattern; this repo is the reference implementation, not something strangers sync verbatim.
+
+## Non-goals
+
+- No multi-page docs site, no SSG framework. One static page.
+- No new skills or CLI features. Release what exists.
+- No rewrite of README; the landing complements it.
+- No template repo; fork-and-edit of this repo is the documented path.
+
+## Milestones
+
+### M-pre: remove omx
+
+omx (oh-my-codex) is retired from the machine and from repo surfaces.
+
+Acceptance:
+- `omx uninstall` has been run; `command -v omx` exits non-zero (manual removal of the binary is allowed after uninstall).
+- `grep -rn --include="*.md" "omx" README.md skills/ plugins/dotagents/skills/` returns no hits (history/experimental notes may stay).
+- `codex exec "echo ok"` works on the cleaned setup.
+
+### M0: history audit, then flip to public
+
+HUMAN GATE before the flip.
+
+Steps:
+1. Run gitleaks (or trufflehog) over the full history: `gitleaks detect --source . --log-opts="--all"`. Zero findings, or each finding explicitly waived by the user.
+2. Produce `tmp/audit-history.md`: every file ever committed (`git log --all --name-only --format= | sort -u | grep -v '^$'`), flagging personal-data candidates (jobs pipeline, telegram/linkedin configs, memory hooks, names/emails beyond git identity).
+3. HUMAN GATE: user reviews the audit summary and approves the flip.
+4. After approval: `gh repo edit --visibility public`.
+
+Acceptance:
+- Secret scan clean (or waivers recorded in the audit file).
+- `gh repo view --json visibility` returns `PUBLIC`.
+- `tmp/audit-history.md` exists locally and is NOT committed (`tmp/` is gitignored).
+
+### M1: verify all 8 harness integrations
+
+Two-tier bar. Tier 1 (all 8: Claude Code, Codex, Droid, Hermes, Pi/OMP, OpenCode, OpenClaw, Amp): scripted, deterministic. Managed harnesses get sync/native-location/status assertions; compatibility-only harnesses get documented-limitation assertions. Tier 2 (harnesses with a headless mode: Claude Code, Codex, Droid, Hermes): a live skill invocation.
+
+Steps:
+1. Write `tmp/verify-harnesses.sh` (not committed): for managed harnesses, run `dotagents sync`, assert skills/MCP land in the harness's native location, and confirm `dotagents status` reports clean; for compatibility-only harnesses, assert the README documents the limitation instead of requiring a managed native surface.
+2. Tier 2 live runs, one cheap prompt each, asserting the agent can see a synced skill:
+   - `claude -p "list available skills, one per line"` contains `tech-search`
+   - `codex exec "list available skills"` contains `tech-search`
+   - `droid exec` equivalent
+   - `hermes -p` equivalent
+3. Claude Code plugin path verified separately from sync: `claude plugin` marketplace add/install flow works against the public repo (after M0), or against the local path pre-flip.
+4. Record results in `tmp/verify-results.md` with exact commands and pass/fail per harness.
+
+Acceptance:
+- `tmp/verify-harnesses.sh` exits 0 covering all 8 harnesses.
+- 4 live-run transcripts captured in `tmp/verify-results.md`, each showing a synced skill visible to the agent. A transcript captured by an authenticated external session (e.g. the user's Claude Code) counts; the executor must not re-run checks that require auth its environment does not have — treat the recorded transcript as the evidence.
+- `dotagents status` clean globally.
+- Any harness that cannot pass gets a fix or a documented limitation in README (compatibility table), not silence.
+
+### M2: landing page
+
+One static page, hand-written HTML/CSS in `docs/site/index.html` (style may echo `docs/harness-map.html`). No build step, no framework.
+
+Content:
+- What dotagents is (tool + canonical-repo pattern, this repo as living reference).
+- Three happy-path scenarios with exact commands:
+  1. Claude Code plugin: marketplace add + install.
+  2. Your own `~/.agents`: fork this repo (or start clean), install CLI, `dotagents setup`, `dotagents sync`.
+  3. Add a skill once, it appears in every harness: create `skills/foo/SKILL.md`, `dotagents sync`, show it visible in 2+ harnesses.
+- Comparison table reused from README (dotagents vs skillshare/vsync/agents-cli).
+- Link to repo and releases.
+- Footer easter egg: one small muted line linking to https://yourconscience.github.io/mushroom/ with a 🍄 icon and a short caption (e.g. "agent-built in one evening"). Keep it footer-only; do not feature it in the main content.
+
+Copy process: draft -> humanizer pass (skill is synced and available) -> HUMAN GATE: user reviews and approves the final copy before deploy. Do not deploy unreviewed copy.
+
+Acceptance:
+- `docs/site/index.html` exists, self-contained (no external JS deps; system fonts or one hosted font max).
+- Every command shown on the page has been executed verbatim during M1/M2 and worked.
+- User approval recorded (gate passed) before M3.
+
+### M3: deploy and release
+
+Steps:
+1. GitHub Pages via Actions workflow deploying `docs/site/` on push to main.
+2. Tag a release `vX.Y.Z` (next semver from existing tags); goreleaser workflow already attaches binaries.
+3. README: add the Pages URL at the top.
+
+Acceptance:
+- Pages URL returns 200 and renders the landing.
+- Release exists with macOS/Linux amd64+arm64 binaries attached.
+- `go install github.com/yourconscience/dotagents/cmd/dotagents@latest` works from a clean environment (now that the repo is public).
+
+## Constraints
+
+- All repo changes on feature branches, PRs into main; `/pr-triage` discipline applies. No worktrees.
+- Commit messages: short single line, no co-author trailers.
+- Do not commit `tmp/` artifacts, audit outputs, or verification transcripts.
+- Quota awareness: heavy execution on Codex (OpenAI Pro); Claude only for humanizer pass and reviews.
+- Anything irreversible (visibility flip, tag push, Pages enable) happens only at its HUMAN GATE or explicitly listed acceptance step.
+
+## Risks / open questions
+
+- History audit may surface personal data that requires history rewrite before flip; if so, stop and present options (filter-repo vs clean mirror) instead of proceeding.
+- OpenClaw/Amp/OpenCode local installs may be stale or broken; budget for "documented limitation" outcome rather than deep debugging.
+- Plugin marketplace flow against a just-flipped repo may need GitHub cache time; retry before declaring failure.
+
+## Codebase notes
+
+- CLI: `cmd/dotagents`; release pipeline: `.github/workflows/release.yml` (goreleaser on `v*` tags).
+- CI: `.github/workflows/ci.yml` (build-and-lint matrix + skill-frontmatter).
+- Existing static-page style reference: `docs/harness-map.html`.
+- Skills inventory and categories: README "Skills" section, `docs/reports/skill-categorization.md`.
+- Plugin marketplace manifest: `.claude-plugin/marketplace.json` (portable skills list).
+
+## Outcome / Deviations
+
+(fill after implementation)

plugins/dotagents/skills/cmux/SKILL.md

@@ -120,7 +120,7 @@ cmux send --workspace workspace:N --surface surface:N "droid"
 cmux send-key --workspace workspace:N --surface surface:N "Enter"
 ```
 
-cmux-specific launchers include `cmux codex-teams`, `cmux claude-teams`, `cmux hermes`, `cmux omx`, and `cmux omo`. Use the normal agent CLI directly when you need exact command control.
+cmux-specific launchers include `cmux codex-teams`, `cmux claude-teams`, and `cmux hermes`. Use the normal agent CLI directly when you need exact command control.
 
 ## Hooks
 

skills/cmux/SKILL.md

@@ -122,7 +122,7 @@ cmux send --workspace workspace:N --surface surface:N "droid"
 cmux send-key --workspace workspace:N --surface surface:N "Enter"
 ```
 
-cmux-specific launchers include `cmux codex-teams`, `cmux claude-teams`, `cmux hermes`, `cmux omx`, and `cmux omo`. Use the normal agent CLI directly when you need exact command control.
+cmux-specific launchers include `cmux codex-teams`, `cmux claude-teams`, and `cmux hermes`. Use the normal agent CLI directly when you need exact command control.
 
 ## Hooks