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 skills add 20alexl/claude-engram --skill engramgit clone --depth 1 https://github.com/20alexl/claude-engramWrote 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/skills/20alexl/claude-engram/engram)<a href="https://agentmods.dev/skills/20alexl/claude-engram/engram"><img src="https://agentmods.dev/badge/skills/20alexl/claude-engram/engram/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/20alexl/claude-engram/engram"><img src="https://agentmods.dev/badge/skills/20alexl/claude-engram/engram.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Memory Poisoning · line 27 Skill injects content designed to persist in agent memory or context across interactions. Persistent injection can alter agent behavior long after the initial interaction.Fix: Do not allow untrusted input to persist in agent memory or context. Validate all content before storing and implement memory isolation between sessions.
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 | $0.00070 | $0.04026 |
| Opus 5.5 | $0.00028 | $0.01610 |
| Sonnet 5.5 | $0.00014 | $0.00805 |
| Haiku 4.5 | $0.00007 | $0.00403 |
Grade A, and why
engram 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 1mo 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 — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Claude Engram — Quick Reference
Slash commands (/engram <subcommand>)
The text after /engram is the subcommand. Do exactly this, nothing more:
run <goal>— the loop is Claude Code's own/goal; engram brackets it. A goal is set only by the person typing/goal: you cannot type a slash command and no tool call sets one. So reply with exactly one line for the person to type,/goal <the goal, written as something your own output can demonstrate; add "or stop after N turns" if a bound is wanted>, and one sentence: from the next stop engram brackets the goal (autonomy mode on: three no-effect strikes halt every tool, ask-first rules refuse their commands, alerts go out, a turn cap then the halt, the goal stamped into every checkpoint, the run report at the end). Then stop; do not start working until the goal is set. When a plan of yours should run unattended, do the same: hand the person the/goalline, never a loop of your own.stop— a goal ends only from the person's keyboard: say "type/goal clear" (a met goal clears itself). If the run is halted, alsopython -m claude_engram.hooks.stall release <session>.- For cron and overnight, outside any session:
python -m claude_engram.run --headless --goal "<goal>"(the launcher; it puts the/goalline in the prompt itself;--dry-runprints the plan). status [<session>]—python -m claude_engram.hooks.stall status <session>(strikes, halted, alerts) plussession_mine(run_report, session_id=<session>)when the run has ended. Without a session id, list.engram/runs/*.manifest.jsonnewest first and use the newest.release <session>— a person is lifting a halt:python -m claude_engram.hooks.stall release <session>, then say tools are allowed again and strikes are reset. Only on the user's word.report [<session>]—session_mine(run_report, session_id=<session>)and summarize: outcome, turns, cost, stalls, compliance matches, alerts, how it ended.- Anything else, or no subcommand: this quick reference.
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.
- 1mo ago Changed · +23 lines · +29 tokens per session 316312e54118
- 1mo ago First seen · 110 lines · 41 tokens per session scan A 9c146c1c1bde
engram is a skill published in the GitHub repository 20alexl/claude-engram (16 stars, last pushed 1mo ago), licensed MIT. It adds 70 tokens to every session and 4,026 once invoked, about $0.0003 per session on Opus 5.5. 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 2026-08-30.
Other skills, from other repositories
zime-memory
A personal memory system that stores text, images, files, and secrets in separate collections. It can find stored items by meaning rather than only by exact words.
coding-agents-hooks-authoring
To author, register, and test Rosetta hooks, add a SemanticKind, or debug a hook that won't fire.
memorix-sessions
Use when resuming work, preparing handoff context, binding an HTTP control-plane project, or deciding whether sessionstart is useful.
frontmcp-extensibility
Use when extending FrontMCP beyond the core SDK by integrating external npm packages, libraries, or third-party services into providers and tools. Covers VectoriaDB for in-memory semantic and vector search (ML-based embeddings or TF-IDF keyword engines, with persistence) and the tamper-evident, hash-chained skill…
memory-audit-pattern-extraction
A method for investigating repeated mistakes by comparing related memories and checking whether an earlier reminder failed. It looks at where the reminder was stored, when it was created, and whether it was strong enough to prevent the mistake.
frontmcp-testing
Use for anything about testing FrontMCP servers: writing or running unit, integration, and E2E tests and reaching the 95%+ coverage bar. Covers Jest setup and coverage gating; unit-testing a ToolContext execute() with mock context, inputs, and Zod schema validation; testing resources and prompts; in-memory testing via…