sync-skill-to-manager

name	sync-skill-to-manager
description	Sync a locally-developed OpenCode skill to the skill-manager npm package and (if private) the private-skills GitHub repo. Handles per-skill version bumps, public/private classification, build verification, and conventional-commit-style git push. Auto-publish to npm is handled downstream by nano-step/shared-workflows@v1 when the push to master lands. Use this skill whenever the user says 'sync skill', 'publish skill', 'push skill to manager', '/sync-skill-to-manager <name>', or asks to release/distribute a skill they just edited.
compatibility	OpenCode with gh CLI authenticated as kokorolx. (npm publish is handled by the auto-publish workflow on master, not by this skill.)
metadata	{"author":"Sisyphus","version":"2.0.0"}

sync-skill-to-manager

Distribute a local OpenCode skill to its publishing channels with one command. Picks up the source skill from the user's project or global config, classifies it as public or private, copies to the right repo(s), bumps per-skill versions, builds, commits, and pushes.

npm publish is NOT this skill's job anymore. The nano-step/shared-workflows@v1 publish-stable reusable workflow on nano-step/skill-manager's master branch handles npm publish automatically when this skill's push lands — including bumping skill-manager's own package.json version, generating CHANGELOG.md, tagging, and creating the GitHub Release. This skill stops at git push.

TL;DR

/sync-skill-to-manager <skill-name>            # sync + commit + push (auto-publish takes over on master)
/sync-skill-to-manager <skill-name> --dry-run  # preview only
/sync-skill-to-manager <skill-name> --no-push  # local commit only (manual push later)

Hard-coded paths (this skill is opinionated about layout)

Var	Value
SKILL_MANAGER_REPO	/Users/tamlh/workspaces/self/AI/Tools/skill-manager
PRIVATE_SKILLS_REPO	/Users/tamlh/workspaces/self/AI/Tools/private-skills
PRIVATE_CATALOG	${SKILL_MANAGER_REPO}/private-catalog.json

If those paths don't exist, abort with a clear error — don't try to "find" them.

Source skill resolution

1. Project: ${PWD}/.opencode/skills/<name>/SKILL.md (if it exists)
2. Global: ~/.config/opencode/skills/<name>/SKILL.md
3. If both exist → ask the user which to use
4. If neither exists → abort

The "source skill" is the directory that contains SKILL.md. We always copy the whole directory (references/, assets/, scripts/, checklists/, CHANGELOG.md, etc.).

Public vs private classification

Read ${PRIVATE_CATALOG} once. Then:

Condition	Classification
<name> IS in private-catalog.json	private
<name> is NOT in private-catalog.json AND ${SKILL_MANAGER_REPO}/skills/<name>/ exists	public
<name> exists in neither place (brand new)	ASK USER: public or private?

Routing rules (privacy-respecting)

Source	Destination	What gets bundled in npm
Public skill	${SKILL_MANAGER_REPO}/skills/<name>/	YES — full skill files
Private skill	${PRIVATE_SKILLS_REPO}/skills/<name>/ AND private-catalog.json entry	NO files — only the catalog entry. Users with a token fetch from GitHub at install time.

Privacy leak fix: skill-manager's installer prefers local skills/ over the GitHub private fetch. If a private skill is physically present in ${SKILL_MANAGER_REPO}/skills/, it will install without auth. The sync MUST detect this and offer to remove the leaked copy:

⚠️  '<name>' is private but a copy exists at ${SKILL_MANAGER_REPO}/skills/<name>/.
   This bypasses authentication. Remove it? (yes/no)

Only proceed with the sync after the user decides.

Workflow

Pre-flight runs the helper script scripts/sync.sh for the heavy lifting. The orchestrator (this skill) handles user prompts, classification decisions, and final summary. The helper handles file mirroring, build verification, and git operations. No npm operations.

Step 1 — Pre-flight checks

Run scripts/preflight.sh <name>. It exits non-zero with a specific message if any check fails:

1. ${SKILL_MANAGER_REPO} and ${PRIVATE_SKILLS_REPO} exist
2. Source skill directory found
3. Source SKILL.md exists and has parseable YAML frontmatter
4. Source skill.json exists and is valid JSON with name, version, description
5. gh auth status shows kokorolx as active
6. Token-leak scan on source: grep -rE 'gho_[A-Za-z0-9]{30,}|ghp_[A-Za-z0-9]{30,}|AKIA[0-9A-Z]{16}' — abort if matches found
7. git -C ${SKILL_MANAGER_REPO} status --porcelain is empty (or only matches the files we're about to touch)
8. git -C ${PRIVATE_SKILLS_REPO} status --porcelain is empty (only checked if private)

(npm whoami is no longer checked — this skill does not publish to npm. That's handled by the auto-publish workflow on the server side using the org-level NPM_TOKEN.)

If any fails: report the exact failure and stop. Do not auto-fix.

Step 2 — Classify and confirm

After preflight:

1. Check private-catalog.json for <name>
2. Check ${SKILL_MANAGER_REPO}/skills/<name>/ and ${PRIVATE_SKILLS_REPO}/skills/<name>/
3. Resolve classification per the table above
4. If brand-new, ask: "New skill detected. Publish as public or private? (public/private)"
5. If private leak detected, ask: "Remove leaked copy at ${SKILL_MANAGER_REPO}/skills/? (yes/no)"
6. Print a plan to the user:

   📦 sync plan for '<name>'
     Classification: <public|private>
     Source:    <resolved-source-path>  (v<source-version>)
     Targets:
       • <target-path-1>  (currently v<remote-version>)
       • <target-path-2>  (catalog entry only)
     Version action: <use-as-is | bump-patch | abort-source-behind>
     Commits:    <count> across <repo-count> repo(s)
     Push:       <yes|no>
     npm publish: (handled by auto-publish workflow on master push)
   Proceed? (yes/no)
   

Step 3 — Version handling

Source-of-truth for version: source skill's skill.json (skill-manager reads this).

Source vs remote skill.json version	Action
Source > remote	Use source as-is. Copy SKILL.md frontmatter version too.
Source == remote	Auto-bump patch in source (and propagate). Print "auto-bumped v → v".
Source < remote	Abort with: "Source v is behind remote v. Pull or set source version manually."

If the version is bumped, also update:

• SKILL.md YAML frontmatter metadata.version
• skill.json version
• private-catalog.json entry's version (private only)

Use scripts/version-bump.sh for the patch bump (semver-aware).

Step 4 — Brand-new skill: README + catalog updates

If brand-new:

• Public: build the table row from skill.json name + description (truncated to one line). Insert into ${SKILL_MANAGER_REPO}/README.md's "Public Skills" table, alphabetized. Show diff. Ask confirm.
• Private: append entry to private-catalog.json (alphabetized) AND insert into the "Private Skills" table in README. Show diff. Ask confirm.

If user rejects the diff → still copy files but skip the README update (with a printed reminder).

Step 5 — Copy files

public:
  rsync -a --delete <source>/ ${SKILL_MANAGER_REPO}/skills/<name>/
private:
  rsync -a --delete <source>/ ${PRIVATE_SKILLS_REPO}/skills/<name>/
  # If leak removal confirmed:
  rm -rf ${SKILL_MANAGER_REPO}/skills/<name>

--delete ensures removed source files are also removed from target (avoids stale files lingering across syncs).

Do NOT copy: .git/, node_modules/, .checkpoints/, .DS_Store. Use --exclude flags.

Step 6 — Build skill-manager (smoke test)

cd ${SKILL_MANAGER_REPO}
npm run build

If build fails → abort. Do NOT commit broken state.

This step exists as a local smoke test — it catches TypeScript errors before pushing broken code to master. The auto-publish workflow on master will also build (via prepublishOnly), but failing fast locally is friendlier.

skill-manager's package.json version is NOT touched by this skill. The auto-publish workflow inspects the conventional-commit message and patches/minors/majors the version itself. Adding our own bump here would cause double-bumps.

Step 7 — Commit

Use scripts/commit.sh which runs (per repo) with conventional-commit messages:

skill-manager commit message templates:

• New public skill: feat(<name>): add v<X.Y.Z>
• Updated public skill: feat(<name>): sync v<old> → v<new> (or chore(<name>): sync v<X> if patch only)
• New private skill (catalog only): feat(<name>): add to private catalog v<X.Y.Z>
• Updated private (catalog version bump only): chore(<name>): bump catalog to v<X.Y.Z>
• Privacy leak removal: fix: remove leaked private skill <name> from public bundle

These can be combined into a single commit when they belong together.

Why the conventional-commit type matters now: The auto-publish workflow translates types into semver bumps — feat: → minor bump on skill-manager, fix:/chore:/docs: → patch bump, anything with ! or BREAKING CHANGE → major bump. Pick the type that reflects the impact on @nano-step/skill-manager users.

private-skills commit message:

• New: feat(<name>): add v<X.Y.Z>
• Update: feat(<name>): sync v<old> → v<new> (or chore if patch)

(The private-skills repo does not yet have its own auto-publish workflow — commit type there is informational only.)

NEVER add Co-authored-by, Signed-off-by, or AI attribution trailers (per workspace AGENTS.md).

Step 8 — Push

If --no-push not set:

git -C ${SKILL_MANAGER_REPO} push
git -C ${PRIVATE_SKILLS_REPO} push   # only if private

The skill-manager remote is HTTPS+token. The private-skills remote uses both SSH (fetch) and HTTPS+token (push) — git push will use the push URL. Both rely on the embedded token or gh auth setup-git. If push fails on auth → STOP and surface the error verbatim.

As soon as the push to master lands, GitHub Actions fires nano-step/skill-manager's Publish Stable workflow (calls nano-step/shared-workflows@v1). That workflow takes over: it inspects the commit, picks the semver bump, updates package.json, regenerates CHANGELOG.md, commits with [skip ci], tags vX.Y.Z, runs npm publish --tag latest, and creates a GitHub Release. You are done from this skill's perspective.

Step 9 — Summary

Print:

✅ sync complete for '<name>'

  Classification: <public|private>
  Source version: v<X.Y.Z>

  Commits:
    • skill-manager:  <sha-short> <subject>
    • private-skills: <sha-short> <subject>   (if applicable)

  Pushed:
    • https://github.com/nano-step/skill-manager/commit/<sha>
    • https://github.com/nano-step/private-skills/commit/<sha>   (if applicable)

  Auto-publish: triggered on push to master.
    Watch: https://github.com/nano-step/skill-manager/actions
    Once the run completes, install with:
      npx @nano-step/skill-manager update <name>

Flags

Flag	Effect
--dry-run	Print the full plan and exit. No writes, no git.
--no-push	Commit locally but don't push. (Auto-publish doesn't fire until you push.)
--force	Skip dirty-state check. Still requires explicit user yes confirmation.
--source <path>	Override source skill location (skip resolution).
--classify public|private	Override classification (required for brand-new skills).
--remove-leak	Required when syncing a private skill that has a leaked copy in skill-manager/skills/. Removes the leaked copy as part of the commit.

--no-publish was removed in v2.0.0 — this skill no longer publishes to npm. Auto-publish is now handled by nano-step/shared-workflows@v1 on nano-step/skill-manager's master branch.

Error messages (be specific)

Failure	Message
Source not found	Source skill '<name>' not found in .opencode/skills or ~/.config/opencode/skills
Both sources exist	(interactive prompt)
Source version behind	Source v<X> is behind remote v<Y>. Pull latest or set source version manually.
Token leak	Token-like pattern detected in <file>:<line>. Refusing to sync. Remove the secret first.
Dirty repo	<repo> has uncommitted changes:\n<status output>\nCommit, stash, or pass --force.
gh auth wrong account	Expected gh active account 'kokorolx', got '<user>'. Run: gh auth switch -u kokorolx
Build failed	npm run build failed in skill-manager. Output:\n<output>\nNot committing.
Push failed	git push to <remote> failed:\n<output>\nManual intervention required.
Privacy leak detected	(interactive prompt)

Anti-patterns (this skill MUST NOT do)

• Push without explicit user invocation of the skill (running the skill IS the consent — but --no-push overrides)
• Run npm publish — that's the auto-publish workflow's job. This skill must never call npm publish, npm version, or modify skill-manager's top-level package.json version field.
• Add AI attribution / Co-authored-by to commits
• Commit private skill files into ${SKILL_MANAGER_REPO}/skills/
• Bump per-skill (inside skills/) major or minor version automatically — only patch
• Skip the build step before committing (it's a smoke test for TS errors)
• Use git push --force ever
• Edit unrelated files in either repo

Scripts

The orchestrator delegates the mechanical work to small bash scripts shipped alongside this skill:

• scripts/sync.sh <name> [flags] — main entry point; coordinates the others
• scripts/preflight.sh <name> — all pre-flight checks (exits non-zero on failure)
• scripts/version-bump.sh <type> <skill-json-path> — patch-only semver bump
• scripts/commit.sh <repo> <message> — conventional commit with no AI trailers

These are kept short and composable so the orchestrator can call them step-by-step and intercept on failure. See scripts/README.md for invocation details.