skill-open-sourcer
Skill787 starsChanged 4 months agoMITView on GitHub
Audit, package, add, verify, and publish local Agent Skills through
- Reads credentials
In this repository20 other files it gives its agents.
- AGENTS.mdAGENTS.md
- codex-cli-model-bridgeSkill
- codex-doctorSkill
- codex-external-handoffSkill
- codex-handoffSkill
- codex-image-genSkill
- codex-model-routing-teamSkill
- codex-skill-adminSkill
- codex-theme-studioSkill
- enterprise-clone-builderSkill
- gpt56-sol-pro-consultSkill
- html-expressSkill
- leadbookSkill
- light-plan-and-workSkill
- scholar-stickersSkill
- web-clipperSkill
- wechat-article-searchSkill
- wechat-stylerSkill
- workbuddy-cli-model-bridgeSkill
- wxmp-article-harvesterSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
16 rules. About 0 are common, 6 are distinctive, and 10 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
10
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Very specific: line 20
Put the complete agent payload in skills/<name>/, human docs in docs/skills/<name>/, and release notes in docs/changelogs/<name>.md.
- Distinctive: line 23
Stop on secrets, private paths, client data, unpublished proprietary material, unclear asset ownership, or an unverified canonical remote.
- Distinctive: line 25
Read Portfolio mode, Registry contract, Release package, and README design before writing.
- Distinctive: line 31
When a Skill release changes Registry inventory or versions, or descriptions used by an existing WorkBuddy catalog, refresh and verify the …
- Very specific: line 37
Follow the canonical repository's Git synchronization rules.
- Very specific: line 79
When README assets are generated by the Portfolio, regenerate them and run the asset tests before accepting the diff.
- Very specific: line 81
Verify local npx skills discovery, then run one fail-fast isolated copy install from the canonical repository:
- Very specific: line 98
Verify the remote Portfolio listing and isolated install, then create the canonical Tag and launch copy.
- Distinctive: line 113
If blocked, return the file paths, reasons, sanitization steps, and safest next action.
- Distinctive: line 130
Never force-push, rewrite published Tags, create a standalone repository, or treat missing evidence as passed.
- Very specific: line 134
Every Skill lives at skills/<skill-name>/, including a single-file Skill.
- Very specific: line 135
Keep SKILL.md lean; put detailed agent guidance in references/ only when needed.
- Very specific: line 138
Put human docs in docs/skills/<skill-name>/ and README visuals in its assets/readme/ directory.
- Very specific: line 139
Update registry/skills.json, root README catalogs, and deterministic asset generators together.
- Very specific: line 152
Never use npx skills add <source> --help as a help probe.
- Distinctive: line 156
After all gates pass, explain the repeated problem, what the Skill automates, the canonical install command, one concrete example request, …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Skill Open Sourcer
Publish every public Skill from the single canonical repository: https://github.com/zjp1997720/zhijian-skills.
Hard publishing boundary
- A
SKILL.mdpath or Skill directory is an import candidate, not a request to create a repository. - Put the complete agent payload in
skills/<name>/, human docs indocs/skills/<name>/, and release notes indocs/changelogs/<name>.md. - Register the Skill in
registry/skills.json. Never addmirror,mirror_tag, standalone repository, redirect workflow, or mirror-export metadata. - Publish commits, per-Skill Tags, and install instructions only from
zjp1997720/zhijian-skills. - Stop on secrets, private paths, client data, unpublished proprietary material, unclear asset ownership, or an unverified canonical remote.
Read Portfolio mode, Registry contract, Release package, and README design before writing.
WorkBuddy marketplace branch
For WorkBuddy custom-source adaptation, marketplace maintenance, or an install-success/cold-start-missing failure, read WorkBuddy marketplace before editing. Follow its completion contract for adapter-only tasks; use the full release workflow below when Skill payloads also change.
When a Skill release changes Registry inventory or versions, or descriptions used by an existing WorkBuddy catalog, refresh and verify the generated marketplace before freezing the release plan. This branch preserves the publishing, source-ownership, and authorization boundaries above.
Workflow
- Resolve the input to a real Skill directory. Accept a
SKILL.md, its containing directory, or a symlink from a local Skill root. - Resolve the canonical checkout from the current repository,
ZHIJIAN_SKILLS_REPO, or~/Documents/GitHub/zhijian-skills. Verifyoriginresolves tozjp1997720/zhijian-skills; never guess another destination. - Follow the canonical repository's Git synchronization rules. Run
git_sync_guard.py check --repo <zhijian-skills>before changing or publishing from an existing checkout; aneeds-syncmarker is a hard stop. Then run:
SKILL_OPEN_SOURCER_DIR="<zhijian-skills>/skills/skill-open-sourcer"
python3 "$SKILL_OPEN_SOURCER_DIR/scripts/check_release_env.py" \
--repo-dir <zhijian-skills> --check-npx-skills
python3 "$SKILL_OPEN_SOURCER_DIR/scripts/scan_skill_release.py" \
/path/to/skill-or-SKILL.md- Inspect
SKILL.mdand directly referencedagents/,references/,scripts/, andassets/. Copy only the complete, sanitized install payload intoskills/<name>/; preserve required license and third-party notices. - Write the eight-field release story from README design, choose
clean-docorproof-led, then create or update bilingual docs, Changelog, Registry record, Portfolio catalog, and project-native visuals. A proof-led Hero must use a composition derived from the Skill's real mechanism or output; changing only the title and motif inside a shared layout fails review. Never expose a root-levelSKILL.mdor add human README files inside the Skill payload. - Choose the version with version-contract.md. Use only the canonical Tag
<skill>/v<version>. - Freeze a release plan with the narrowest selector. For a governed Skill, its
manifest.jsonmust declaretrust_reportandoutput_quality_scorecardas files inside the install payload; both files must exist and be tracked atHEAD. Planning fails before candidate refs are created when either baseline is undeclared, missing, untracked, or escapes the payload. Use--skill <name>for one Skill; reserve--alland optional--excludefor an intentional multi-Skill wave:
python3 "$SKILL_OPEN_SOURCER_DIR/scripts/release_portfolio.py" plan \
--repo <zhijian-skills> --source-checkout <original-checkout> \
--skill <name> --dry-run \
--plan-out /tmp/<name>-release-plan.json
python3 "$SKILL_OPEN_SOURCER_DIR/scripts/release_portfolio.py" verify \
--plan /tmp/<name>-release-plan.json--skill and --all are mutually exclusive. Repeat --skill only when the user explicitly authorizes a small multi-Skill release. Do not combine --skill with --exclude.
The plan records the live origin/main SHA with git ls-remote. verify queries it again and fails when it changed. When --repo is a temporary clean clone, --source-checkout must identify the original checkout; planning marks that checkout release-in-progress and installs a pre-commit guard so it cannot create new commits from the stale base.
- Validate the Skill, declared tests, full Portfolio, documentation, and repository contracts:
python3 "$SKILL_OPEN_SOURCER_DIR/scripts/portfolio.py" \
validate-skill <zhijian-skills>/skills/<name>
python3 "$SKILL_OPEN_SOURCER_DIR/scripts/portfolio.py" \
audit --repo <zhijian-skills> --strict
python3 "$SKILL_OPEN_SOURCER_DIR/scripts/audit_release_readme.py" \
--repository-root <zhijian-skills> \
<zhijian-skills>/docs/skills/<name>/README.md \
<zhijian-skills>/docs/skills/<name>/README.zh-CN.md --strict
python3 -m unittest discover -s <zhijian-skills>/tests -vWhen README assets are generated by the Portfolio, regenerate them and run the asset tests before accepting the diff. Render every changed Hero at 900px and 360px; deterministic checks do not replace visual inspection.
- Verify local
npx skillsdiscovery, then run one fail-fast isolated copy install from the canonical repository:
python3 "$SKILL_OPEN_SOURCER_DIR/scripts/verify_isolated_install.py" \
--repo <zhijian-skills> --skill <name>The verifier isolates HOME, uses copy mode, and compares SHA-256 manifests byte-for-byte. It returns non-zero immediately when preflight, installation, materialization, missing/extra files, or content comparison fails; do not append a success-printing command that can mask its exit code. Listing success alone is insufficient. Use top-level help only: npx --no-install skills --help. Never run npx skills add <source> --help; skills CLI 1.5.x may perform a real installation because <source> is already a valid add request.
10. Commit only on a short-lived branch created from the recorded remote SHA. Immediately before the remote wave, run verify again, push the branch, and merge it through a PR into protected main; Agents never push directly to main. After merge, read the live remote SHA and record the verified transition:
python3 "$SKILL_OPEN_SOURCER_DIR/scripts/release_portfolio.py" record-step \
--plan /tmp/<name>-release-plan.json --skill <name> \
--step canonical-pushed --remote-sha <merged-origin-main-sha>This step verifies that remote main contains the planned source and upgrades a temporary-clone source marker to needs-sync. Never call gh repo create or publish to <owner>/<skill-name>.
11. Verify the remote Portfolio listing and isolated install, then create the canonical Tag and launch copy. Clear a source checkout marker only after that checkout is clean and its local HEAD exactly equals the verified remote SHA.
Output contract
Return:
- canonical Skill URL under
zjp1997720/zhijian-skills - the canonical commit, version, and
<skill>/v<version>Tag - install command using
npx skills add zjp1997720/zhijian-skills - Skill validation, Portfolio audit, remote listing, and isolated-install results
- README presentation tier and deliberate visual assets
- the eight-field release story, unique composition choice, and rendered desktop/mobile inspection result
- safety summary and residual assumptions
- at least one X/Twitter launch post, plus a Chinese version for Chinese users
If blocked, return the file paths, reasons, sanitization steps, and safest next action.
Safety rules
Hard stop on:
- API keys, tokens, private keys, cookies, credentials,
.env, or auth config - personal absolute paths, local vault paths, or machine-only caches
- client data, private URLs, unpublished prompts, or assets without redistribution rights
- escaping symlinks, databases, logs, browser profiles,
.DS_Store, dependency caches, or unexplained binaries - a dirty or unverified canonical repository that cannot be reconciled safely
- a
release-in-progressorneeds-synccheckout marker - remote
mainchanging after the release plan is frozen - a governed
trust_reportoroutput_quality_scorecardthat is undeclared, missing, outside the payload, or not tracked atHEAD
Literal credential assignments are blockers. Runtime transformations such as URL-encoding a token variable are not literal secrets and must not be blocked by the generic assignment detector; high-confidence provider key patterns remain blocking.
Never force-push, rewrite published Tags, create a standalone repository, or treat missing evidence as passed.
Packaging rules
- Every Skill lives at
skills/<skill-name>/, including a single-file Skill. - Keep
SKILL.mdlean; put detailed agent guidance inreferences/only when needed. - Preserve
agents/openai.yaml; create it when absent with a$<skill-name>default prompt. - Include deterministic scripts and licensed agent-facing assets only.
- Put human docs in
docs/skills/<skill-name>/and README visuals in itsassets/readme/directory. - Update
registry/skills.json, root README catalogs, and deterministic asset generators together. Every generated proof-led Hero needs a unique, stabledata-compositionvalue and a matching project-specific generator branch; Portfolio tests must reject duplicate composition IDs.
Publication verification
npx --no-install skills --help
npx --no-install skills add zjp1997720/zhijian-skills --list
python3 scripts/verify_isolated_install.py \
--repo <zhijian-skills> --skill <skill-name> \
--install-source zjp1997720/zhijian-skillsThe verifier uses the Portfolio's pinned skills CLI and returns non-zero for every failed phase. For a cold check where the pinned package is not installed, install dependencies in the canonical checkout before running it; never improvise a shell chain whose final command can hide an earlier failure.
Never use npx skills add <source> --help as a help probe. It may perform a real installation and create .agents/ plus skills-lock.json in the current directory.
Launch copy
After all gates pass, explain the repeated problem, what the Skill automates, the canonical install command, one concrete example request, and any relevant safety boundary. Never link a standalone Skill repository.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

