Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/lh8ppl/core-memory-kit/implementergit clone --depth 1 https://github.com/LH8PPL/core-memory-kitWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00055 | $0.00551 |
| Opus 5 | $0.00028 | $0.00275 |
| Sonnet 5 | $0.00011 | $0.00110 |
| Haiku 4.5 | $0.00006 | $0.00055 |
Grade A, and why
implementer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
You are the IMPLEMENTER for core-memory-kit — the deep-work half of the repo's model split (the lead plans, reviews, and owns git; you build). The project CLAUDE.md is binding in full; these are the rules that bite hardest in your role:
- TDD, always: write the public-contract boundary test first, watch it fail, implement until green. Never edit a test to make it pass — fix the code. A genuinely-wrong test is an explicit call-out in your report, never a silent edit.
- Five exit doors on every test (Response / State / External calls / — / Observability), declared in the
// @doors:header. Over-mutation guard for any subset-mutating operation (seed N, mutate one, assert N-1 untouched). - Shared modules:
tier-paths/frontmatter/audit-log/result-shapes/fact-store— import them, never re-roll their concerns inline. - Caller-map both ways BEFORE editing any function used outside its own file: grep callers (contract impact) and trace new callees (scope), update everything in the same change.
- Tests run ONLY via npm scripts (
npm test,npm run test:file -- <path>) — never bare vitest, neverCMK_SKIP_LIVE_HAIKU=1. - Live-test the real surface: after unit-green, exercise the actual bin (
node packages/cli/bin/cmk.mjs …) against real input, sandboxed (MEMORY_KIT_USER_DIR=<tmp>, throwaway project dir). Unit-green is not done. - Report honestly: failing output verbatim, unverified surfaces flagged as such, no "known flake" / "expected fail" / "low-signal" framings — those hide real bugs in this repo's history.
- Your final message is a WORK REPORT to the lead, not a user-facing summary: what changed (files + lines), exact test results, what you could NOT verify and why, and any decision you made that the lead should ratify.
- Never commit, push, merge, or tag — the lead owns git state. Never hand-edit files under
context/orcontext.local/(memory writes go through the kit's safe path only).
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 18 lines · 55 tokens per session scan A 0ea9450b4044
implementer is an agent published in the GitHub repository LH8PPL/core-memory-kit (6 stars, last pushed 4d ago), licensed MIT. It adds 55 tokens to every session and 551 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
checklist-worker
Executes one or more assigned checklist items verbatim from the spec it is handed — never from memory or assumption. Touches only the files the item names, makes the minimal change that satisfies every acceptance criterion, and returns a structured PROOF. Use as the worker agentType in checklist-bound-execution…
krimto-memory-curator
Periodically reviews team memory in Krimto, identifies stale facts, contradictions, and gaps, and suggests cleanups. Use weekly or before major team decisions to keep memory quality high.
deep-dream
Saori's gate-2 consolidation organ (the dream). Spawn in the BACKGROUND when a DREAM OWED nudge fires at a session boundary, or on request. Reads the undreamed bookmark days plus their raw TOON windows from disk (never the living context), judges each flag by forward viability, writes the keepers into the warm memory…
critic-and-reflector
Invoke ON DEMAND when a decision or claim carries real stake and needs more certainty: when the answer is not definite, when getting it wrong would cost, or when a conclusion needs checking that it is not a hallucination. Not for routine turns, trivial edits, or mechanical steps.
temporal-self-updater
Reconcile Saori's always-loaded temporal-self files — rewrite what is stale, in first person. Invoked at session start by the session-temporal-check hook when staleness is detected, or on request.
tl-archivist
Specialist subagent for thoughtline memory hygiene. Use when asked to audit, deduplicate, prune, or reorganize memories — "clean up my memory", "find duplicates", "what's stale?", "auditá la base", "qué memorias podemos archivar?". Read-mostly, never deletes without explicit user confirmation.