agentleFS
roronoazoroshao369roronoazoroshao369/vibe-coding-os/skills/memory/memory-search/SKILL.md

memory-search

Skill1 starsChanged 5 months agoMITView on GitHub

Find candidate memory entries for a task using staged search, then return only relevant, cited, confidence-labeled context. Use before planning, debugging, reviewing, or implementing when prior sessions, decisions, commands, or constraints may change the answer. Applied as original wording from Supermemory-inspired retrieval/search design and claude-mem-inspired progressive disclosure: hybrid keyword/path/semantic search, threshold tuning, and reranking guidance without requiring vector infrastructure. Not applied: Chroma, SQLite, embedding services, hosted search, SDK clients, or copied upstream text. Tìm bộ nhớ theo từng lớp: tìm…

11 other projects use 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. 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 7 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use before planning, debugging, reviewing, or implementing when prior sessions, decisions, commands, or constraints may change the answer.

    line 20
  2. Distinctive:

    Search terms, entities, files, time window, and scope filters.

    line 26
  3. Distinctive:

    Search broad indexes or summaries first.

    line 31
  4. Distinctive:

    If nothing useful is found, say so and proceed without inventing memory.

    line 36

“Common” is measured: kept by at least 6 in 10 of the 7 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

Skill: Memory Search

Purpose

Find candidate memory entries for a task using staged search, then return only relevant, cited, confidence-labeled context.

When to use

Use before planning, debugging, reviewing, or implementing when prior sessions, decisions, commands, or constraints may change the answer.

Inputs

  • Current task and search question.
  • Available local memory, project memory, session summaries, handoffs, and cited observations.
  • Search terms, entities, files, time window, and scope filters.

Workflow

  1. Convert the task into 2-5 search queries: entity, file/path, decision, error, and workflow terms.
  2. Search broad indexes or summaries first; avoid loading raw transcripts by default.
  3. Inspect only promising entries, following citation IDs or source paths.
  4. Tune retrieval without binding to vector infrastructure: blend keyword/path hits for exact names with semantic summaries for fuzzy intent; start with a conservative relevance threshold, lower it only when recall is clearly missing, and rerank by source fit, recency, confidence, and contradiction risk.
  5. Deduplicate by source and keep the newest non-contradicted entry unless history matters.
  6. Label every returned fact with source, confidence, staleness, and contradictions.
  7. If nothing useful is found, say so and proceed without inventing memory.

Outputs

  • Focused memory search report.
  • Relevant citations and rejected/stale matches.
  • Open questions or missing-memory note.

Failure modes

  • Injecting every matching entry into context.
  • Treating semantic similarity as evidence.
  • Dropping source/citation labels.
  • Ignoring stale or contradicted entries.

Verification checklist

  • Search used more than one term or path when ambiguity existed.
  • Hybrid keyword/path/semantic search was tuned only as far as the task required.
  • Threshold and rerank choices are explainable without provider-specific claims.
  • Raw context was loaded only after summary hits justified it.
  • Returned context is cited, scoped, and confidence-labeled.
  • Irrelevant/stale matches were excluded or marked.

Applied / Not Applied

Applied as original wording from Supermemory-inspired retrieval/search design and claude-mem-inspired progressive disclosure: hybrid keyword/path/semantic search, threshold tuning, and reranking guidance without requiring vector infrastructure. Not applied: Chroma, SQLite, embedding services, hosted search, SDK clients, or copied upstream text.

Ghi chú tiếng Việt

Tìm bộ nhớ theo từng lớp: tìm rộng trước, đọc chi tiết sau. Chỉ đưa vào ngữ cảnh phần thật sự liên quan, có nguồn, độ tin cậy, và nhãn cũ/mâu thuẫn.

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.