agentleFS
Sign inStart free
IgorGanapolskyIgorGanapolsky/ThumbGate/.agents/skills/cobble-hot-store-compare-not-clone/SKILL.md

cobble-hot-store-compare-not-clone

Skill29 starsChanged 7 months agoMITView on GitHub

CobbleDB (Perplexity 2026-09) is a batch-read hot store, not a ThumbGate clone. Steal the three-plane FORMAT (durable state / batched delivery / query-time MultiGet + hedge + hot subset) onto existing lesson rails; never vendor RocksDB, YTsaurus, Pillar, or Lorry. Slash: /cobble-hot-store-compare-not-clone.

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    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:

    Keep the hot subset to promoted/matchable records.

    line 69

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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

CobbleDB — compare, do not clone

Goal

Produce fail-closed honesty for whom: ThumbGate agents writing feedback or serving lessons — so processing never contends with PreToolUse reads, batch lookups hedge slow sources, and only promoted/matchable records occupy the hot path.

Constraints

NEVERALWAYS
Clone CobbleDB / Pillar / Lorry / RocksDB / YTsaurusMap planes onto existing lesson rails
Write processing output onto hybrid-feedback-contextDurable log → feedback-to-memory → retrieval
Sequential per-key hot reads of a batchbatchGetPrepared (partition + hedge)
Dump raw feedback-log.jsonl into retrievalpromoted_matchable (or fresh_high_value) subset
Overwrite embedding/chunk versionsCoexist representations
Dual-edit DIRTY lesson-graph PR #3650Leave graph layer alone
Quote Perplexity 5× / 20% as oursMeasure ThumbGate rails with perf-budget-check
Hero Continuity / net-new storage SKUECI: existing retrieval rails only

HARD fail closed. REFUSE SKU clones.

Reference

  • https://www.perplexity.ai/hub/blog/cobbledb
  • scripts/cobble-hot-store-split.js
  • scripts/lesson-retrieval.js · scripts/feedback-to-memory.js · scripts/memory-vs-rag-route.js
  • docs/agents/cobble-hot-store-split.md
  • /high-roi-steal-and-finish · /eci-thumbgate-ip-wall

Examples (show, don't tell)

Weak: Summarize CobbleDB and add a RocksDB cluster class.

Gold:

$ npx thumbgate cobble-hot-store-split --json
ok: true
status: ready
$ node --test tests/cobble-hot-store-split.test.js
# processing → hot write → coupled_processing_hot_write deny

Procedures

npx thumbgate cobble-hot-store-split --json
npx thumbgate cobble-hot-store-split --map-only
npx thumbgate cobble-hot-store-split --trace=tests/fixtures/cobble-hot-store-split-coupled.json --json
npx thumbgate cobble-hot-store-split --keys=lesson-1,lesson-2 --json
npm run test:cobble-hot-store-split
  1. Classify every write as durable, delivery, or hot.
  2. Deny processing → hot coupling.
  3. Require batched MultiGet (+ hedge) for ≥3 serving keys.
  4. Keep the hot subset to promoted/matchable records.
  5. Refuse --clone-cobbledb.

Rubric

  • gold default (this repo) → ok=true, status=ready
  • --trace coupled fixture → ok=false, coupled_processing_hot_write
  • --clone-cobbledb → ok=false, cobble_clone_refused
  • sequential ≥3-key hot read → sequential_hot_batch
  • unbounded hot subset → unbounded_hot_subset
  • doctor: npm run test:cobble-hot-store-split PASS
  • evidence: command output in the same turn

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