agentleFS
pantheon-orgpantheon-org/tekhne/skills/agentic-harness/vault-search/SKILL.md

vault-search

Skill10 starsChanged 39 days agoMITView on GitHub

Search persistent memory for relevant context — including architectural decisions, resolved bugs, and implementation patterns — when the user asks about past decisions, when debugging a recurring issue, when uncertain whether something was already decided, or before re-implementing something that may have been done before. Use when the user says "do you remember…", "what did we decide about…", or "have we solved this before".

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. 0 are common, 11 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 11

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Search before you implement, design, or debug — not after.

    line 19
  2. Distinctive:

    When search returns empty, say so honestly.

    line 21
  3. Distinctive:

    Before starting a non-trivial task: search for related prior decisions

    line 25
  4. Distinctive:

    When the user asks "do you remember…" or "what did we decide about…"

    line 26
  5. Distinctive:

    Before proposing an approach: check for existing constraints or patterns

    line 27
  6. Distinctive:

    When a bug appears familiar: search for past workarounds

    line 28
  7. Very specific:

    Use noun phrases, not questions: "auth token expiry" not "how does auth token expiry work?"

    line 40
  8. Very specific:

    Use --project to narrow scope when working in a known project context

    line 41
  9. Very specific:

    If the first query returns nothing, broaden: remove specific version numbers or error codes and retry with a higher-level term (e.g.

    line 42
  10. Distinctive:

    If the search exits with "No results.":

    line 47
  11. Distinctive:

    If still empty, tell the user no prior memory exists on this topic

    line 49
  12. Distinctive:

    Never invent or guess at a prior decision — proceed fresh and offer to capture the new decision

    line 50
  13. Distinctive:

    Never fabricate a memory: when search returns empty — report "No results found" and proceed without invented context

    line 73
  14. Distinctive:

    Search memory first; do not re-derive what was already decided.

    line 78

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “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

vault-search

Retrieve relevant memories from the vault using hybrid BM25 + vector search.

Mindset

Search before you implement, design, or debug — not after. A 2-second search that surfaces a prior constraint saves hours of work in the wrong direction. When search returns empty, say so honestly; never fabricate a memory.

When to use

  • Before starting a non-trivial task: search for related prior decisions
  • When the user asks "do you remember…" or "what did we decide about…"
  • Before proposing an approach: check for existing constraints or patterns
  • When a bug appears familiar: search for past workarounds

How to use

vault-cli search "<query>" [--top-k <n>] [--project <id>]

Exits 1 with "No results." if nothing is found.

Query construction

  • Use noun phrases, not questions: "auth token expiry" not "how does auth token expiry work?"
  • Use --project to narrow scope when working in a known project context
  • If the first query returns nothing, broaden: remove specific version numbers or error codes and retry with a higher-level term (e.g. "postgres connection" instead of "ECONNREFUSED 5432")

Handling no results

If the search exits with "No results.":

  1. Retry with a broader synonym query
  2. If still empty, tell the user no prior memory exists on this topic
  3. Never invent or guess at a prior decision — proceed fresh and offer to capture the new decision

Examples

vault-cli search "authentication approach"
vault-cli search "database migration strategy" --project my-app
vault-cli search "recurring TypeScript error" --top-k 3
vault-cli search "postgres connection refused" --project api

Output format

Each result shows:

1. [category] summary (date)
   tier: episodic | strength: 0.87 | score: 0.0312
   content preview...

Never

  • Never skip search before a non-trivial implementation task — prior constraints may invalidate your approach before you write a line
  • Never fabricate a memory when search returns empty — report "No results found" and proceed without invented context
  • Never search with a full sentence or question — BM25 scores individual terms; verbose queries dilute signal

Mindset

Search memory first; do not re-derive what was already decided. Treat retrieval as cheap and re-work as expensive. Know when not to: trivial, self-evident steps do not need a memory lookup.

Anti-Patterns

NEVER re-implement something without searching memory first

  • WHY: past decisions, constraints, and fixes are often already recorded; re-deriving them wastes effort and risks contradicting a prior decision.
  • BAD: writing a new approach from scratch for a recurring problem.
  • GOOD: vault-cli search "<topic>" first, then build on what returns.

NEVER search with the user's literal phrasing when concrete nouns work better

  • WHY: memory is indexed on concepts, not surface wording.
  • BAD: searching the whole question verbatim.
  • GOOD: extract the key nouns and search those.

ALWAYS treat an empty result as meaningful

  • WHY: a miss suggests the decision is genuinely new and worth capturing.

References

  • Vault Search: Query Tips - practical query construction guidance.

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.