Llm wiki github action

LLM-Wiki GitHub Action

Goal

Create safe GitHub Actions workflows for LLM-Wiki validation and maintenance.

When to use

• The user wants a scheduled wiki-lint or wiki-health report published as a workflow artifact.
• The user wants PR-based automation for generated docs or maintenance changes and wants to avoid direct pushes to main.
• The user needs Agent Skills frontmatter validation (or other repo-specific validation scripts) wired into CI.
• The user wants to add or enable one of the starter evals, ingestion, publish, or security workflows.
• The user is deciding what GitHub Actions permissions or triggers a wiki-maintenance job should have.

Starter templates

Use these local templates when available in a full repository checkout. In a single-skill install, recreate the same workflow shapes from this list:

• llm-wiki-evals.github-actions.yml for retrieval, grounding, promptfoo and wiki-health checks.
• llm-wiki-ingestion.github-actions.yml for source manifests, ingestion fidelity, parser security and retrieval smoke checks.
• llm-wiki-publish.github-actions.yml for export profile validation, redaction/security checks, artifacts and optional Pages publish.
• llm-wiki-security.github-actions.yml for secrets, dependency, OSV, Semgrep, red-team and export-policy checks.

The starter workflow schedules are commented out by default. Enable cron only after workflow_dispatch has run successfully and the repository owner accepts the recurring cost and external calls.

Inputs

• Repository path.
• Desired automation: validation, lint, scheduled report, maintenance PR, skill validation.
• Permission constraints.
• Whether Actions should write files, open PRs, or only upload reports.

Procedure

1. Choose automation mode

Mode	Use when
report-only	Early adoption or sensitive repos.
artifact report	Reports should not change the repo.
PR-based update	Generated output needs human review.
validation	Validate schemas, links, skills and frontmatter.

Default to report-only or PR-based updates. Avoid direct pushes to main.

2. Define triggers

Start with manual and pull request triggers. Add a schedule only after the first manual run succeeds. Keep copied starter schedule: blocks commented until that happens.

on:
  pull_request:
  workflow_dispatch:

3. Set permissions narrowly

Use the least privilege required by the chosen mode. Read-only jobs should not request write permissions. PR-creation jobs should be clearly separated from validation jobs.

4. Add validation jobs

Possible jobs:

• Agent Skills frontmatter validation for every skills/*/SKILL.md;
• repository-specific validation scripts when the target repo already defines them;
• frontmatter schema validation;
• broken link scan;
• wiki-lint report generation;
• stale page report;
• provenance gap report.

5. Add PR workflow only when safe

For PR-based updates:

• create a branch;
• commit reports or generated docs;
• open PR with clear summary;
• request review from owners;
• never merge automatically.

6. Copy starter workflows deliberately

When the user asks for ready-made workflow files, copy only the relevant starters:

Need	Starter
evals and red-team	llm-wiki-evals.github-actions.yml
ingestion pipeline checks	llm-wiki-ingestion.github-actions.yml
publish/export pipeline	llm-wiki-publish.github-actions.yml
security scans	llm-wiki-security.github-actions.yml

Rename each copied file to .github/workflows/<purpose>.yml, replace placeholder commands, and keep permissions minimal. Re-verify that the pinned action versions and commands in a copied starter are still current for the target repository's toolchain before enabling its trigger.

Output

## GitHub Actions plan

## Workflow files created or proposed

## Permissions

## Triggers

## Safety model

## Manual verification steps

Safety gates

• Do not grant broad write permissions without justification.
• Do not configure direct pushes to main for generated knowledge.
• Do not run external model calls on private content without explicit approval.
• Do not hide generated changes outside pull-request review.