Centralize skillify command list in SKILL.md via build-step generator

[efenocchi]

Context

PR #129 introduced SKILLIFY_COMMANDS in src/cli/skillify-spec.ts — the single source of truth for the hivemind skillify ... command list. As part of #129 we centralized all TypeScript-side consumers of that list:

• 4 per-agent SessionStart inject blocks (claude-code, codex, cursor, hermes) via renderSkillifyCommands()
• The pi inline mirror in pi/extension-source/hivemind.ts (guarded by a drift test in tests/pi/skillify-spec-drift.test.ts)
• hivemind --help top-level output in src/cli/index.ts (centralized in feat(skillify): one-shot local skill mining for non-auth users (mine-local) #129 as part of the kaghni feedback)
• hivemind skillify --help usage() in src/commands/skillify.ts (centralized in feat(skillify): one-shot local skill mining for non-auth users (mine-local) #129)

What remains hand-typed are the 3 shipped SKILL.md files:

• claude-code/skills/hivemind-memory/SKILL.md — ## Skill Management (skillify) section
• codex/skills/deeplake-memory/SKILL.md — same block
• openclaw/skills/SKILL.md — same block (pre-dates feat(skillify): one-shot local skill mining for non-auth users (mine-local) #129, on main since April)

These can't be solved the same way as the TS callers because SKILL.md is shipped Markdown — it can't import TypeScript at runtime. We need a build-step generator.

Proposed approach

1. Markers in each SKILL.md — wrap the command block in HTML-comment markers so the generator can locate it deterministically without ambiguity:

   <!-- SKILLIFY_COMMANDS_START -->
   - `hivemind skillify` — show current scope...
   ...
   <!-- SKILLIFY_COMMANDS_END -->

2. Generator script — add scripts/sync-md-commands.mjs that:

   • Imports the compiled SKILLIFY_COMMANDS from bundle/cli.js (or dist/skillify-spec.js)
   • Renders the block in the same bullet format the SKILL.md files use today
   • Rewrites the content between <!-- SKILLIFY_COMMANDS_START --> / END in all 3 files

3. Wire into npm run build — append node scripts/sync-md-commands.mjs to the build pipeline so the markdown can't drift in CI.

4. Drift detection test — under tests/, mirror the structure of tests/pi/skillify-spec-drift.test.ts: read each SKILL.md, parse the block between markers, assert it matches what renderSkillifyCommands() would produce. Fails CI if a developer edits the markdown by hand and skips the build.

5. Same pattern for ORG_COMMANDS (optional, can be a separate sub-task) — the org/auth command list (hivemind login, whoami, org list, …) is hand-typed in the same 3 SKILL.md files plus the per-agent SessionStart hooks. Extracting an ORG_COMMANDS constant and applying the same generator closes the symmetric debt.

Why a separate PR

• Scope: feat(skillify): one-shot local skill mining for non-auth users (mine-local) #129 ships the hivemind skillify mine-local user-facing feature. Wiring a Markdown generator + drift test would tangle the build pipeline with the feature work and expand the review surface significantly.
• Risk surface: A scripts/sync-md-commands.mjs + npm run build change touches CI and can break the release flow. Worth landing on its own branch where it's the focus.
• Independence: This change has zero behavior impact on the mine-local feature — it's pure documentation upkeep.

Acceptance criteria

• All 3 SKILL.md files have the skillify command block generated from SKILLIFY_COMMANDS
• npm run build regenerates the blocks deterministically
• Drift test fails if a SKILL.md block is edited by hand and not regenerated
• Optionally: same applied to org/auth commands via ORG_COMMANDS

Reference

• Original observation: PR #129 review comment by @kaghni