garden-recall
Skill4 starsChanged 6 months agoMITView on GitHub
Search the garden vault for notes relevant to a query and surface them with citations. Use when the user asks about past decisions ("what did I decide about X"), past work ("what's the state of Y"), or anything where the vault might already hold the answer. Greps + reads files under ~/garden/.
In this repository4 other files it gives its agents.
- garden-bootstrapSkill
- garden-captureSkill
- gardenerSkill
- garden-voiceSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 6 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
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
- Distinctive: line 35
Read the top 3–5 files in full: Only the ones the summary pass flagged as worth the tokens.
- Distinctive: line 37
Follow links one hop: when top hits reference unread notes that look on-topic.
- Very specific: line 59
If a candidate has no summary: field, fall back to reading the first 20 lines.
- Distinctive: line 65
Before answering any question where the vault might already hold context: silent recall, then synthesize
- Distinctive: line 69
Don't read every match: top-ranked few only.
- Distinctive: line 70
Don't guess if the vault has nothing: say so plainly: "Nothing in the vault on that yet."
- Distinctive: line 71
Don't write to the vault from this skill.
≈ “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.
Files
garden-recall
Pulls relevant notes from ~/garden/ for a query. Used by the SessionStart hook automatically; can also be invoked explicitly.
How
- Identify search terms from the query. Include synonyms and related concepts.
- Grep the vault for matches:
grep -ril "<term>" ~/garden --include="*.md" | head -20 - Skim summaries first (cheap pass). For each candidate, read just the frontmatter to get title +
summary:+updated::
Notes ship with a one-sentencefor f in <candidate-files>; do awk '/^---$/{n++; if(n==2) exit} n==1' "$f" echo "FILE: $f" echo "---" donesummary:field for exactly this purpose. Decide which 3–5 are worth opening in full based on summaries alone. - Rank candidates by relevance:
- Summary directly answers the query > summary tangentially related
- Exact match in title (frontmatter or H1) > body match
- Recent
updated:> older - Files in
decisions/andprojects/> genericnotes/
- Read the top 3–5 files in full. Only the ones the summary pass flagged as worth the tokens.
Follow links one hop when top hits reference unread notes that look on-topic. Use the typed edges in frontmatter to prune:
supersedes: [X]→ if the query is about current state, skip X (it's been replaced). If the query is historical, X is exactly what to read.depends-on: [X]→ follow X only if the query is about prerequisites or root cause.contradicts: [X]→ always show both sides; surface the tension to the user.derived-from: [X]→ follow X only if the user wants source material or provenance.part-of: [X]→ X is the parent; read it for broader context. Sibling splits (other notes with the samepart-of) may be relevant too.- Plain
[[wiki-links]]in the body → skim summary first, then decide.
For reverse-direction lookups ("what supersedes this?" / "what depends on this?"), grep:
grep -rl "^supersedes:.*<note-name>" ~/garden --include="*.md"- Synthesize for the user with citations (use the note's
summary:as the gist when it's accurate):From the vault: - [[<note-name>]] (updated YYYY-MM-DD): <summary or refined gist> - … <synthesized answer>
If a candidate has no summary: field, fall back to reading the first 20 lines. The gardener will backfill missing summaries on its next run; recall stays read-only.
When to use
- User asks "what did we decide about X"
- User asks about state of a project, person, or topic
- Before answering any question where the vault might already hold context: silent recall, then synthesize
Don't
- Don't read every match: top-ranked few only.
- Don't guess if the vault has nothing: say so plainly: "Nothing in the vault on that yet."
- Don't write to the vault from this skill. Recall is read-only.
Speed
For broad recall (many candidate files), use:
grep -ril "<term>" ~/garden --include="*.md" | xargs -I{} sh -c 'echo "=== {} ==="; head -20 "{}"' | head -200to get a fast overview before deep-reading.
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
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.

