AI agent skill · by @levnikolaevich

Ln 61 Repository Publisher

Commits, pushes and remotely verifies authorized repository changes; does not create releases.

Archived 2026-10-23 · Source updated 2026-09-16 · License: MIT. Check the original repository for the latest version.

SKILL.md instructions

# Repository Publisher

**Goal:** Publish only changes the user has authorized, then verify the result from the remote source.

**Execution contract:** The checklist defines completion. Track each item internally as `PENDING`, `PROVEN` with evidence, `CLEARED` with evidence its condition is absent, or `UNPROVEN` with a gap; reading, delegation, or tool failure is not proof. Reconcile after each section. Before returning, resolve all `PENDING`, count only `PROVEN` and `CLEARED`, and apply verdict and approval rules to every gap.
Preserve intent, scope, and existing authorization. Continue authorized work; ask only for consequential unresolved choices or required external approval. Scale depth to material risk without skipping checks. Preserve dependency and safety order; otherwise choose an appropriate verification method.
Accept equivalent user or repository evidence; no other skill, named artifact, or complete lifecycle is required. Preserve source requirement and decision IDs. Bind reused evidence to relevant source versions, dirty changes, configuration, and environment; invalidate only affected claims.
On continuation, reconcile task, authorization, current state, and unresolved evidence. For long work, return a compact continuation record or update an already authorized artifact; read-only skills do not persist it. Distinguish artifact readiness, verified behavior, and external-action authority.
Prepare authorized work before required approval. If blocked by an instruction, cite its exact source and unresolved boundary; do not invent approval gates from caution.


## Tool Routing

| Need | Preferred capability | Fallback |
|---|---|---|
| Scope, staging, history, and synchronization | Native Git CLI | Stop if equivalent Git evidence is unavailable |
| Repository validation | Project-native commands and installed validators | Documented manual checks |
| Remote branch and CI state | Authenticated hosting CLI or connector | Git remote evidence plus direct workflow URLs |
| Clean-source verification | Temporary clone and isolated host configuration when distribution surfaces changed | Remote raw files plus `BLOCKED` for required install evidence |
| Static-site verification | Deployment workflow plus direct HTTP request | Hosting API deployment state |
| Current marketplace behavior | Official host documentation | Mark assumptions and avoid destructive retries |

Prefer compact Git output first, then open the full diff for files that will be staged. Never pipe commands in a way that hides the failing exit code.

Use hosting APIs for remote facts and Git for repository facts. Local distribution or deployment state cannot prove that a remote consumer update works.

## Checklist

### Establish scope and evidence

- [ ] Confirm the user explicitly requested a commit and push and identify the intended branch.
- [ ] Read repository instructions, release rules, and the current branch policy before mutation.
- [ ] Check `git status -sb`, staged and unstaged diffs, untracked files, remotes, and recent commit style.
- [ ] Identify unrelated user changes; do not stage them without explicit whole-worktree authorization.
- [ ] Inspect deletions and generated files as carefully as edited text.
- [ ] Check whether behavior, installation commands, catalogs, layout, or the public site require matching documentation updates.
- [ ] Do not change versions, tags, or release metadata during an ordinary publication unless the request or repository policy explicitly includes them.
- [ ] For marketplace edits, confirm every stable identifier is unchanged unless an intentional migration was approved.
- [ ] When multiple host catalogs exist and repository policy requires parity, confirm they contain the same distribution units in the required order.
- [ ] When metadata is duplicated across manifests or catalogs, confirm descriptions and source paths agree with the canonical source.

### Validation Routing

- [ ] Discover and run repository-native validation commands before generic checks.
- [ ] For changed skills, run repository-required or host-native skill validators, or perform their documented manual fallback.
- [ ] For changed plugins or packages, run repository-required validators, or perform their documented manual fallback.
- [ ] Run every host-native strict validator whose distribution surface exists and changed or is required by repository policy.
- [ ] Run only the catalog parity, manifest parsing, stale-reference, local-link, and whitespace checks required by the repository and affected surfaces.
- [ ] Run only relevant product tests; do not invent a heavyweight release gate absent from repository policy.
- [ ] Stop before commit on a confirmed failing required check unless the user explicitly accepts the failure.
- [ ] Record skipped checks with the exact missing dependency or environment.

### Synchronization and Commit

- [ ] Fetch the target remote and compare local HEAD with the remote branch before committing.
- [ ] If behind or diverged, inspect both sides and reconcile within the authorized branch workflow while preserving user changes; stop for unresolved semantic conflicts or a required history rewrite. Never force-push implicitly.
- [ ] Stage explicit paths when the worktree is mixed; use whole-worktree staging only when the user approved all changes.
- [ ] Review the cached diff and diffstat after staging; ensure required checks cover this exact tree, rerunning only checks invalidated by synchronization or subsequent edits.
- [ ] Exclude secrets, local caches, temporary artifacts, and unintended credentials from the staged set without deleting user-owned files; if an intended change contains a secret, block that publication and report redacted evidence.
- [ ] Match the repository's commit-message convention and summarize the entire staged change.
- [ ] Preserve configured commit signing and attribution policy; do not invent contributor identities or disable required signing.
- [ ] Create the commit and capture its full SHA.
- [ ] Push to the authorized branch without changing branch protections or using force.

### Remote Verification

- [ ] Verify remote publication refers to the exact checked source state and exclude unverified later changes; repository publication alone does not prove a release or deployment.
- [ ] Verify the published commit on the authorized remote through Git or the hosting API; use a second source only if identity or synchronization is uncertain.
- [ ] Track required CI for the pushed commit with bounded waits and direct run URLs. If execution cannot continue waiting, report pending state as `PARTIAL`; do not equate pending with success.
- [ ] If the static site changed, wait for deployment and verify live content with a cache-busting request.
- [ ] If installation or marketplace content changed, clone an authorized consumer-accessible remote into a clean temporary directory and validate the affected distribution surface.
- [ ] When install or update behavior changed, test at least one affected package or plugin from its documented distribution source in isolated host configuration.
- [ ] When stable distribution identifiers or versions exist, verify the installed artifact resolves under the expected identifier and version.
- [ ] Keep temporary host configuration isolated from the user's active settings and remove it safely afterward.
- [ ] Recheck local worktree and HEAD against the published commit. If the remote advanced concurrently, verify that it still contains the published commit and report both SHAs; do not overwrite newer work to restore equality.

### Safety Gates

- [ ] Never expose authentication tokens or credential values in output.
- [ ] Never create a release, tag, package publication, discussion, or pull request unless explicitly requested.
- [ ] Never delete remote branches or alter the default branch as a side effect.
- [ ] Return `BLOCKED` rather than bypassing branch protection, authentication, or required checks.

## Verdict

- `PUBLISHED` — commit, push, required CI, and applicable remote verification succeeded.
- `PARTIAL` — the push succeeded but a non-destructive remote check is pending or failed.
- `BLOCKED` — publication did not complete because authorization, synchronization, validation, or remote access failed.

## Self-Check

- [ ] **Reconcile before returning.** Check item-level evidence, requirement coverage, contradictions, scope, verdict, and applicable cleanup. Correct the report or authorized artifacts. Reuse valid evidence; do not automatically rescan the repository or rerun successful commands. Repeat checks only for relevant changes, failures, or unresolved evidence. Disclose remaining gaps.

## Output Contract

Report in the user's language, in this order; retain all five fields and state each fact once. Small results may use one line per field; omit empty tables and do not copy linked artifacts:

1. **Result:** Skill-specific verdict and supported outcome.
2. **Scope:** Reviewed/changed scope, exclusions, baseline, and material assumptions.
3. **Evidence:** Skill-specific fields below; distinguish facts, inferences, and unverified claims. Link artifacts; use tables when useful.
4. **Verification:** Checks/results, unavailable evidence, and applicable cleanup/external state.
5. **Completion:** `Checklist: X/Y complete`; `Incomplete: None` or each `UNPROVEN` item's reason, outcome impact, and exact next action; residual risks and required decisions.

**Skill-specific evidence:** Published/excluded paths, branch, full commit SHA, remote URL, validation results, CI/deployment URLs, and applicable clean-source install/update proof. For `PARTIAL`, state whether the pushed commit is safe to leave and the observable event that closes verification. Use `PUBLISHED` only when the remote commit and every applicable required workflow are observed.