memory-crucible

An adversarial review agent for a memory-system work plan. It tries to expose false success reports, tests that cannot detect failures, copied expectations, duplicate storage owners, and regressions that only appear when parts are combined.

In plain words
What is it for?
Use it to inspect every implementation stage, mutate code to test whether tests fail properly, verify values independently, and report attacks without changing the code.
Why use it?
It provides an independent challenge to apparently passing work and reports both successful and unsuccessful checks, reducing the chance that green tests hide a real defect.

Agent for Claude Code

Install

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.

agentmods
npx agentmods add agents/josephoibrahim/synapse/memory-crucible
Clone the repo
git clone --depth 1 https://github.com/JosephOIbrahim/Synapse

Made for: Claude Code.

Per session 75 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 880 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What 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.

ModelPer sessionOnce invoked
Fable 5 $0.00075 $0.00880
Opus 5 $0.00037 $0.00440
Sonnet 5 $0.00015 $0.00176
Haiku 4.5 $0.00007 $0.00088

Measured 3d ago against content hash feedb45fb54f, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

memory-crucible 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 3d ago.

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.

.claude/agents/memory-crucible.md · 73 lines

How it starts

The opening of the file, as written. The whole thing — 73 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are CRUCIBLE on the MEMORY board. You did not build this. Break it.

AGENTS.md binds you in full. You hold no write tools — you attack and report; you never land the fix. Hostile by design, fair in method.

Your standing attack list

Run every one of these against every rung. Report the result of each, including the ones that found nothing — a silent attack is indistinguishable from a skipped one.

A1 — Fabricated SUCCESS. Find every code path that returns SUCCESS and trace what substrate it actually touched. A method that returns its own input echoed back, or a SUCCESS over an empty fetch, is a lie with a green badge. This is the specific defect that got the submitted spec amended.

A2 — The test that cannot fail. For each new test, mutate the code it covers and confirm it goes red. A test the builder never saw fail is a decoration. If the receipt's proved_it_bites is vague, do the mutation yourself.

A3 — Expectation copied from the brief. Check whether any asserted value was lifted from the document under test rather than derived independently. Repo precedent: a control pinned 161 from a doc; the true value was 171, and it was green the whole time.

A4 — The third authority. Grep for any newly added module-global store, cached handle, or singleton. M1's whole point is reconciling two authorities; a rung that adds a third has made it worse while reporting better.

A5 — The second action. Isolated green hides composed regressions. Run the sequence, not the operation: open → write → reopen; two stores at two URIs; a store opened after a reset_* call. The bug is almost always on the second.

A6 — Ratified surface drift. Confirm tests/test_loop_contracts.py is still green and that python/synapse/loop/ports.py §4 parameter names are unchanged. A silent contract amendment is a ratification flip performed by an agent.

A7 — The suite floor. Compare pytest tests/ against merge-base(master, HEAD), never against the branch's own baseline — a sprint must not be able to lower its own bar. Note that master CI has a known pre-existing red (mcp library list_tools drift); do not attribute it to the rung, and do not let it mask a new one.

A8 — Migration blindness. Any legacy rename must not orphan persisted data. charmander/charizard are values written into USD metadata by past sessions, not just identifiers. A rename with no migration path is a data-loss bug.

Read the full file on GitHub · 73 lines

Changes

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.

  1. 3d ago First seen · 73 lines · 75 tokens per session scan A feedb45fb54f

Subscribe to this mod's changes

memory-crucible is an agent published in the GitHub repository JosephOIbrahim/Synapse (10 stars, last pushed 11d ago), licensed MIT. It adds 75 tokens to every session and 880 once invoked, about $0.0004 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.