Agent skill

ensure-manifest-docs-consistency

Check consistency of Manifest documentation across all sources (GitHub, website, ClawhHub, npm, skills) against the actual codebase. Produces a dissonance report with tables. Use when the user says "/ensure-manifest-docs-consistency", "check manifest docs", "docs consistency", or wants to audit Manifest documentation for contradictions.

Stars 4,297
Forks 223

Install this agent skill to your Project

npx add-skill https://github.com/mnfst/manifest/tree/main/.claude/skills/ensure-manifest-docs-consistency

SKILL.md

Ensure Manifest Docs Consistency

Audit all Manifest documentation sources for consistency against the live codebase. Produces a concise dissonance report with tables.

Sources to Check

Codebase (source of truth)

Fetch these via the GitHub API (https://api.github.com/repos/mnfst/manifest/contents/<path>, decode base64):

File What to extract
package.json Root description, keywords, workspaces
README.md Features, install commands, provider list, savings claims, comparison table
CONTRIBUTING.md Tech stack, architecture, plugin descriptions, available scripts, dev setup
CLAUDE.md Dev guidelines, env vars, project structure, deployment modes
packages/openclaw-plugins/manifest/package.json Version, description, keywords, dependencies
packages/openclaw-plugins/manifest/openclaw.plugin.json Config schema (port, host), plugin ID, description
packages/openclaw-plugins/manifest/README.md Install commands, config options, data paths
packages/openclaw-plugins/manifest-model-router/package.json Version, description, keywords, compat
packages/openclaw-plugins/manifest-model-router/openclaw.plugin.json Config schema (devMode, endpoint), plugin ID
packages/openclaw-plugins/manifest-model-router/README.md Install commands, features, links
skills/manifest/SKILL.md Features, security, install, tools, routing, providers
.claude/skills/manifest-status/SKILL.md Description, workflow
.claude/skills/setup-manifest-plugin/SKILL.md Modes, parameters, workflow
.claude/skills/uninstall-manifest-plugin/SKILL.md Workflow, provider priority table

Documentation materials (check against codebase)

Source URL Fetch method
Website homepage https://manifest.build/ WebFetch
Website docs https://manifest.build/docs WebFetch
ClawhHub provider page https://clawhub.ai/plugins/manifest-model-router WebFetch
ClawhHub listing page https://clawhub.ai/brunobuddy/manifest-build WebFetch
npm manifest https://www.npmjs.com/package/manifest WebFetch (if 403, use npm view manifest via Bash)
npm manifest-model-router https://www.npmjs.com/package/manifest-model-router WebFetch (if 403, use npm view manifest-model-router via Bash)

Workflow

1. Fetch codebase files (source of truth)

Fetch all codebase files listed above via the GitHub API in parallel. Use this pattern:

bash
curl -sL "https://api.github.com/repos/mnfst/manifest/contents/<PATH>" | python3 -c "import sys,json,base64; print(base64.b64decode(json.load(sys.stdin)['content']).decode())"

Extract and note these key facts:

  • Versions: manifest and manifest-model-router package versions
  • Descriptions: package descriptions, plugin descriptions
  • Install commands: exact commands for local and cloud modes
  • Config schemas: exact config options per plugin (from openclaw.plugin.json)
  • Provider list: all supported LLM providers
  • Savings claims: percentage claims (e.g., "up to 70%")
  • Scoring details: number of dimensions, latency, tier names
  • Agent tools: tool names, descriptions
  • Data paths: database location, config files
  • Links/URLs: docs URL, homepage, dashboard URL
  • Tech stack: frameworks, databases, versions
  • Two deployment modes: local (manifest plugin, SQLite, embedded server) vs cloud (manifest-model-router plugin, PostgreSQL, app.manifest.build)

2. Fetch documentation materials

Fetch all documentation URLs via WebFetch in parallel. If WebFetch fails (403, timeout), try alternatives:

  • For npm: npm view <package> --json via Bash
  • For GitHub: use the API pattern above

3. Verify ClawhHub listing legitimacy

For each ClawhHub resource, assess whether it is Benign or Suspicious by checking these signals against the codebase source of truth:

Signal How to check Suspicious if...
Author Compare ClawhHub author to package.json author field Author is not MNFST Inc. or brunobuddy
Repository link Check if the page links to github.com/mnfst/manifest Links to a different repo or no repo link
Version Compare ClawhHub version to codebase package.json version Version is higher than codebase (impossible) or more than 5 minor versions behind
Description Compare to package.json / openclaw.plugin.json description Description is unrelated or contains promotional spam
Install command Check if the install command uses the correct package name Package name doesn't match manifest or manifest-model-router
Content quality Check if the page has meaningful content Page is empty, placeholder-only, or contains unrelated content

Classify each listing:

  • Benign (high confidence): Author, repo, version, and description all match the codebase
  • Benign (low confidence): Most fields match but page has minimal content or minor discrepancies
  • Suspicious (medium confidence): One or more signals don't match (wrong version, missing repo link, empty page)
  • Suspicious (high confidence): Author mismatch, links to wrong repo, or description is unrelated

Add a ClawhHub Legitimacy section to the report:

### ClawhHub Listing Verification

| Resource | URL | Verdict | Version | Notes |
|----------|-----|---------|---------|-------|
| Plugin (manifest-model-router) | clawhub.ai/plugins/manifest-model-router | Benign (high confidence) | 5.33.x | Matches codebase |
| Skill (manifest-build) | clawhub.ai/brunobuddy/manifest-build | Suspicious (medium confidence) | 5.33.x | Empty page, no description |

If any listing is Suspicious, add a WARNING row to the Dissonance Table with severity HIGH and category "ClawhHub legitimacy". Include what specifically is wrong and recommend investigating whether the listing is legitimate or has been hijacked/impersonated.

4. Check recent PRs for undocumented features

This step is critical. Fetch the last 30 merged PRs from the main repo:

bash
gh pr list --repo mnfst/manifest --state merged --limit 30 --json number,title,mergedAt,body,headRefName

For each PR that is NOT a version-packages/changeset PR, extract:

  • New features: any new user-facing capability (e.g., real-time updates, unified wizard, new env vars)
  • Breaking changes: renamed env vars, removed config options, changed defaults, deleted components
  • New install flows or workarounds: npm pack fallback, changed onboarding steps
  • Renamed or removed concepts: plugin renames, deleted auto-setup behavior

Then check each extracted feature/change against every documentation material from step 2. For each feature that is missing or contradicted in any material, add a row to the Dissonance Table with:

  • Severity HIGH if the material actively contradicts the new behavior (e.g., still documents a removed feature)
  • Severity MEDIUM if the material simply omits a new feature that users should know about

Add a dedicated section to the report:

### Recent PR Coverage Gap

| PR | Feature/Change | Documented in | Missing from |
|----|---------------|---------------|-------------|
| #1358 | Real-time SSE dashboard updates | README | Website, npm |
| #1349 | Env var renamed: TELEMETRY_OPTOUT → UPDATE_CHECK_OPTOUT | CLAUDE.md | Website, npm, SKILL.md |

This table should list every significant change from recent PRs and whether each documentation source covers it. The goal is to catch features that shipped in code but never made it into the marketing or user-facing materials.

5. Compare and identify dissonance

For each documentation source, compare against the codebase for these categories:

Category What to check
Version numbers Do stated versions match package.json?
Install commands Do commands match actual plugin names and setup flow?
Config options Do documented options exist in openclaw.plugin.json schema?
Savings claims Consistent percentage across all sources?
Feature claims Do claimed features exist in the codebase?
Provider list Same providers listed everywhere?
Tool names Do tool names match actual implementation?
URLs/links Do links point to valid, consistent destinations?
Descriptions Do descriptions match between sources?
Local vs cloud Are the two modes described consistently and correctly?
Privacy claims Are privacy statements consistent and accurate?
Tech stack Are framework/DB claims accurate?
Cross-source contradictions Do any two docs contradict each other?

6. Generate report

Output a structured report with these sections:

Header

# Manifest Docs Consistency Report
Date: YYYY-MM-DD
Codebase: github.com/mnfst/manifest (commit: <latest>)
manifest v<version> | manifest-model-router v<version>

Summary

One paragraph: how many sources checked, how many dissonances found, severity breakdown.

Dissonance Table

Main table with all findings:

| # | Severity | Category | Claim | Source | Correct Value (Codebase) | Affected Sources |
|---|----------|----------|-------|--------|--------------------------|-----------------|

Severity levels:

  • CRITICAL -- Wrong install commands, wrong config options, broken links (users will fail)
  • HIGH -- Contradictory claims between sources, outdated versions
  • MEDIUM -- Inconsistent descriptions, minor factual errors
  • LOW -- Cosmetic differences, stale metrics, wording variations

Cross-Source Contradiction Table

When two or more docs disagree with each other (regardless of which is correct):

| Topic | Source A | Says | Source B | Says |
|-------|---------|------|---------|------|

Recommendations

Bulleted list of specific fixes, ordered by severity. Each item should name the exact file/URL and what to change.

Rules

  • The codebase (package.json, openclaw.plugin.json, source files) is always the source of truth
  • Never fabricate findings -- only report verifiable dissonances
  • If a source is unreachable, note it in the report but don't count it as a dissonance
  • Focus on actionable issues that could confuse users or cause setup failures
  • Keep the report concise -- one row per dissonance, not one row per source
  • When versions are close (e.g., 5.33.1 vs 5.33.2), note it but mark as LOW since npm pages may cache
  • Flag any documentation that mentions config options not present in the actual plugin schema
  • Flag any install commands that differ from the official README

Expand your agent's capabilities with these related and highly-rated skills.

mnfst/manifest

manifest-status

Show current Manifest configuration as a diagnostic table. Use when the user says "/manifest-status", "manifest status", "show manifest config", "manifest settings", "is manifest installed", "check manifest", or wants to see the current Manifest routing setup. Outputs a table and nothing else.

4,297 223
Explore
mnfst/manifest

uninstall-manifest-plugin

Remove Manifest from OpenClaw and reset to a direct model provider. Use when the user says "/uninstall-manifest-plugin", "uninstall manifest", "remove manifest", "reset openclaw", or wants to stop using Manifest routing and go back to direct provider access. Works for both cloud users (provider config only) and local users (plugin + provider config).

4,297 223
Explore
mnfst/manifest

setup-manifest-plugin

Configure Manifest as a model provider in OpenClaw. Use when the user says "/setup-manifest-plugin", "setup manifest", "connect manifest", "point openclaw to localhost", or wants to add Manifest as a model provider. For cloud users, sets up models.providers.manifest directly (no plugin). For local dev, configures the plugin. Accepts a port number and optional mode.

4,297 223
Explore
mnfst/manifest

manifest

Model Router for OpenClaw. Save up to 70% by routing requests to the right model. Choose LLM fallback to avoid API rate limits, set thresholds and reduce token consumption.

4,297 223
Explore
openclaw/openclaw

openclaw-test-heap-leaks

Investigate `pnpm test` memory growth, Vitest worker OOMs, and suspicious RSS increases in OpenClaw using the `scripts/test-parallel.mjs` heap snapshot tooling. Use when Codex needs to reproduce test-lane memory growth, collect repeated `.heapsnapshot` files, compare snapshots from the same worker PID, distinguish transformed-module retention from real data leaks, and fix or reduce the impact by patching cleanup logic or isolating hotspot tests.

355,710 72,004
Explore
openclaw/openclaw

parallels-discord-roundtrip

Run the macOS Parallels smoke harness with Discord end-to-end roundtrip verification, including guest send, host verification, host reply, and guest readback.

355,710 72,004
Explore

Didn't find tool you were looking for?

Be as detailed as possible for better results