agentleFS
Sign inStart free
frankxaifrankxai/Starlight-Intelligence-System/skills/intelligence/hermes-search/SKILL.md

intelligence/hermes-search

Skill10 starsChanged 2 months agoMITView on GitHub

Use when you need to find, retrieve, or research anything across vaults, repos, or the web — executes semantic search with full provenance, deduplicates across sources, flags contradictions, and returns ranked results sourced to the exact vault atom, file, or URL. Default skill for Starlight Hermes; callable by all agents.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 3 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When a search is scoped to a private namespace, Sentinel Claw is notified and the scope is logged.

    line 110
  2. Distinctive:

    Agents: All agents can invoke Hermes Search as a sub-skill.

    line 116
  3. Very specific:

    Vaults: Read-only across all six substrate vaults.

    line 122
  4. Distinctive:

    Every result includes a source (vault name + atom ID, or file path, or URL)

    line 128

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Hermes Search

Find exactly what you need — from vaults, repos, or the web — with full provenance.

When This Skill Activates

  • Keywords: "search", "find me", "where is", "look up", "retrieve", "what do we know", "source check", "provenance", "research", "what did we decide", "trace", "find pattern", "cross-vault", "multi-source"
  • Commands: /search, /research, /find-pattern, /audit-coverage, /trace-provenance, /find-contradiction, /source-check
  • Agents: starlight-hermes (primary), starlight-prime, starlight-sage, starlight-genius, starlight-orchestrator
  • Intents: search, retrieval, research, provenance, audit

What This Skill Does

Hermes Search is the retrieval engine for the SIS substrate. It routes queries to the right source, executes parallel searches, deduplicates and ranks results, flags contradictions, and returns every result with a full provenance chain.

This skill is read-only. It never writes to vaults, repos, or external systems.

Procedures

Procedure 1: Vault Search

1. Decompose query into search terms + intent
2. Route to relevant vaults (all 6 by default, or scoped by context)
3. Execute BM25 full-text + semantic similarity
4. Rank results by: relevance × recency × confidence
5. Deduplicate by content hash
6. Return results with: vault name, atom ID, date, confidence, contributing agent

Procedure 2: Cross-Repo Search

1. Identify repos to search (current + linked via GitHub MCP)
2. Execute code/docs/issue search via GitHub MCP
3. Surface: file path, line number, commit date, author
4. Rank by: relevance × recency × repo authority
5. Return results with full GitHub provenance chain

Procedure 3: Web Research

1. Only when local sources are exhausted or explicitly requested
2. Decompose query into search terms
3. Execute structured search (not raw scrape)
4. Apply credibility filter: prefer primary sources, official docs, peer-reviewed
5. Return ranked results with: URL, date, credibility signal, snippet

Procedure 4: Multi-Source Synthesis

1. Execute vault + repo + web searches in parallel
2. Merge results from all sources
3. Deduplicate across sources (same fact from multiple sources → highest-authority wins)
4. Run contradiction check: do any results conflict?
   - If yes: flag contradiction explicitly. Do NOT silently resolve.
5. Rank merged set by: relevance × source authority × cross-corroboration count
6. Synthesize into a single answer with inline source citations
7. Preserve uncertainty: if confidence is below threshold, say so

Procedure 5: Contradiction Detection

1. Given a set of facts (from vault search or provided by caller)
2. Compare each fact against others for direct contradiction
3. Flag: fact A says X, fact B says NOT-X — source chain for both
4. Do NOT pick a winner
5. Surface to user / requesting agent for resolution
6. Log contradiction in Operational Vault for Memory Claw review

Procedure 6: Coverage Audit

1. Given a topic or domain
2. Search all vaults for existing coverage
3. List what exists (with sources and confidence)
4. List what is absent (known gaps)
5. Rate coverage: sparse / partial / substantial / comprehensive
6. Recommend: which agent + Claw should fill the gaps

Privacy Boundaries

NamespaceDefault scope
Strategic, Technical, Creative, Operational, Wisdom, Horizon✅ Included by default
second-brain/❌ Excluded unless explicitly scoped
health/❌ Excluded unless explicitly scoped
business/❌ Excluded unless explicitly scoped
relational/❌ Excluded unless explicitly scoped

When a search is scoped to a private namespace, Sentinel Claw is notified and the scope is logged.

Integration Points

Agents: All agents can invoke Hermes Search as a sub-skill. Hermes agent is the primary activator but any agent can call it.

Claws: Genius Claw uses Hermes to mine the personal corpus. Reclamation Claw uses Hermes to detect duplicates. Attestation Claw uses Hermes to surface prior attestation records. Memory Claw uses Hermes to avoid duplicate vault writes.

MCP: Uses sis-memory-mcp for vault read, GitHub MCP for cross-repo search, browser/web MCP for research (optional).

Vaults: Read-only across all six substrate vaults. Write path is Memory Claw → sis-memory-mcp.

Quality Criteria

  • Every result includes a source (vault name + atom ID, or file path, or URL)
  • Contradictions are flagged, not silently resolved
  • Confidence is calibrated (not inflated to satisfy the query)
  • Privacy boundaries are respected
  • Results are ranked, not dumped raw
  • The answer distinguishes between "found in vault" (high confidence) and "found on web" (lower confidence)

Read-only · Provenance-first · Contradiction-surfacing · Built on SIP

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.

Check your setup, free