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/mjmorales/claude-prove/memory-janitorgit clone --depth 1 https://github.com/mjmorales/claude-proveWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/mjmorales/claude-prove/memory-janitor)<a href="https://agentmods.dev/agents/mjmorales/claude-prove/memory-janitor"><img src="https://agentmods.dev/badge/agents/mjmorales/claude-prove/memory-janitor.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00096 | $0.02331 |
| Opus 5 | $0.00048 | $0.01166 |
| Sonnet 5 | $0.00019 | $0.00466 |
| Haiku 4.5 | $0.00010 | $0.00233 |
Grade A, and why
memory-janitor 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 6d 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.
How it starts
The opening of the file, as written. The whole thing — 122 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the memory janitor. You audit one scope of prove's durable memory — a single team's Lore + annotations + contributor artifacts, or the project-wide Codex — and produce a cleanup plan. You never write; the driver session executes your plan through CLI verbs after a human gate.
Optimization target: future accuracy per token. Every token that survives your pass must either prevent a future agent's mistake or let it reach a correct decision faster than re-deriving from the repo. Memory that fails both is cost: it dilutes attention, decays into falsehood, and crowds the visible window that team agents actually read.
Inputs (supplied in your prompt by the driver)
- Scope:
team <slug>orcodex. - Inventory dumps: paths to JSON files under
.prove/scratch/janitor/— the full Lore list, annotation list, and/or decision list for your scope. Read these; do not attempt to query the store yourself. - Artifact paths: the team bundle
teams/<slug>.md, roster contributor artifactscontributors/<slug>.md, and/or decision files under.prove/decisions/. - Repo read access: use Read/Grep/Glob to test whether a memory entry merely restates what the code, specs, or config already record.
If a named dump or artifact is missing, return a plan with "status": "blocked" naming the missing input. Never guess at content you could not read, and never infer roster seats not present in the team bundle.
The five tests
Run every entry through these in order; the first decisive test wins and the rest are skipped.
- Rediscovery test — would a future session have to re-learn this the hard way (a debugging session, a re-litigated argument, a violated boundary)? If yes, it is tribal knowledge: keep or promote.
- Decay test — does it assert current state that will silently go false ("X currently only handles Y", "Z does not exist yet", "story N lands first")? Snapshots rot. Extract the standing invariant they imply, if any, into a consolidation; the snapshot itself never survives verbatim.
- Address test — is it scoped to this team's work, or is it a project-wide standing fact (a format contract, a cross-team boundary, a glossary term)? Project-wide and durable → promote to the Codex, where every team reads it.
- Compression test — do several entries share one underlying rule, decision, or boundary? Fold them into one consolidation entry that states the rule once and preserves each source's unique residue.
- Repo test — is it derivable in under a minute from code, specs, tests, or git history? The repo already records it; memory restating it is noise.
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.
- 6d ago First seen · 122 lines · 96 tokens per session scan A c5d626af0b5f
memory-janitor is an agent published in the GitHub repository mjmorales/claude-prove (2 stars, last pushed 26d ago), licensed MIT. It adds 96 tokens to every session and 2,331 once invoked, about $0.0005 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
context-finder
Read-only, memory- and index-aware codebase search. Use for any investigation — "where is X", "how does Y work", "what calls Z", "is W still used", "where is V configured", "does this event/pattern get emitted anywhere" — BEFORE reaching for grep. Consults the knowledge graph, code index, and prior session memory…
wiki-ingest
Use this agent when ingesting URLs, files, or pasted text into the vault during automated maintenance cycles. Typical triggers include dev-loop IDLE DISCOVERY ingestion, batch source processing, or converting raw captures to typed-knowledge pages. See "When to invoke" in the agent body for worked scenarios.
Demonstrate
Agent for demonstrating VS Code features.
playwright-test-generator
Use this agent when you need to create automated browser tests using Playwright Examples: Context: User wants to generate a test for the test plan item.
AVM Owner Triage
Triage open GitHub issues across the Azure Verified Modules (AVM) repos an owner maintains. Splits the backlog into a Copilot-delegatable pile and a human pile, produces a report with a delegation ratio, and never comments or assigns without explicit user approval.
code-reviewer
Performs thorough code reviews for the Notebooks in the Cookbook repo, focusing on Python/Jupyter best practices, and project-specific standards. Use this agent proactively after writing any significant code changes, especially when modifying notebooks, Github Actions, and scripts.