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 FerroxLabs/ijfw --skill ijfw-auto-memorizegit clone --depth 1 https://github.com/FerroxLabs/ijfwWrote 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/ferroxlabs/ijfw/ijfw-auto-memorize)<a href="https://agentmods.dev/skills/ferroxlabs/ijfw/ijfw-auto-memorize"><img src="https://agentmods.dev/badge/skills/ferroxlabs/ijfw/ijfw-auto-memorize/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/ferroxlabs/ijfw/ijfw-auto-memorize"><img src="https://agentmods.dev/badge/skills/ferroxlabs/ijfw/ijfw-auto-memorize.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00037 | $0.00959 |
| Opus 5 | $0.00018 | $0.00479 |
| Sonnet 5 | $0.00007 | $0.00192 |
| Haiku 4.5 | $0.00004 | $0.00096 |
Grade C, and why
ijfw-auto-memorize scanned grade C with 1 finding 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 5d 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.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
> *Stored 3 new memories: pagination-off-by-one fix, user prefers esbuild, stopped repeating rm -rf warnings.* How it starts
The opening of the file, as written. The whole thing — 75 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Fires at session end. Reads deterministic signals captured during the session
and synthesizes structured memories. Nothing leaves the machine unless the user
explicitly configured an API model via IJFW_AUTOMEM_MODEL.
Consent gate (first run only)
Before any synthesis, check .ijfw/.automem-consent:
- If missing: ask the user once: "IJFW can automatically extract lessons (errors hit, fixes applied, preferences you stated) at session end into local memory. OK? (y/n). Reply
y,n, orask(ask again next time)." Write answer as{"consented": true|false, "at": "<iso>"}to.ijfw/.automem-consent. - If
"consented": false: do nothing this session. - If
"consented": true: proceed.
Inputs (all local files)
.ijfw/.session-signals.jsonl-- ERROR/FAIL/Traceback lines captured by the PreToolUse hook (W3.6)..ijfw/.session-feedback.jsonl-- corrections/confirmations/preferences detected by the UserPromptSubmit hook (W3.7)..ijfw/.prompt-check-state-- last turn's intent + vague signals..ijfw/memory/project-journal.md-- existing entries (dedupe against these).- Transcript read via Claude Code's Stop-hook payload (
transcript_path).
Synthesis
For each signal cluster:
- Redact secrets first. Call
redactSecrets()frommcp-server/src/redactor.json every field that came from transcript or tool output. - Cap sizes. Run
applyCapsfrommcp-server/src/caps.js. content ≤4KB, why/how ≤1KB, summary ≤120. - Dedupe. Use BM25 search (
mcp-server/src/search-bm25.js) againstproject-journal.md. If score > 6 against an existing entry, skip (duplicate). - Classify into one of:
pattern-- error→fix recurrence (same error type seen >=2x).decision-- an explicit user choice ("from now on X").preference-- a style/workflow preference ("I prefer Y").observation-- something worth noting, single instance.
- Emit via
ijfw_memory_storeMCP tool with fields:type: one of the abovesummary: single sentence, ≤120 charscontent: the fact + minimal contextwhy: where this came from (e.g., "user said 'don't use X'", or "hit error Y at step Z")how_to_apply: when this should surface in future sessionstags: includeauto-memorizeand the classifier kind (correction,confirmation,preference,rule,error)
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.
- 5d ago First seen · 75 lines · 37 tokens per session scan C 603a850eddb3
ijfw-auto-memorize is a skill published in the GitHub repository FerroxLabs/ijfw (210 stars, last pushed 3d ago), licensed MIT. It adds 37 tokens to every session and 959 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-05.
Other skills, from other repositories
repo-context-ledger
Record every behavior-changing feature addition, fix, and adjustment as durable, evidence-based repository knowledge, then use that ledger to continue accurately across AI windows, tools, Git collaboration, and pull requests. Use the deterministic runtime to route bounded context, isolate private drafts, publish…
memory
A session-memory tool with three layers: agent notes, requirement memories, and overall project memory. The description does not specify how these memories are stored or managed.
summary
A skill that writes a summary of the current conversation to a daily memory file before work ends.
compact-task-memory
Produce a lossy compaction summary of TASKSTATE.md when task memory has grown beyond a useful working size, preserving canonical decisions and recommended next step while dropping stale facts. Distinct from sync-task-state (incremental, append-only, never lossy) and state-reconcile (drift repair, no shrinking). Use…
capture-observation
Append a single observation, question, hypothesis, or concern to TASKSTATE.md as task memory without disrupting in-progress work or requiring a full state sync. Lean append-only; never restructures other artifacts. Use when something surfaces mid-work that should be remembered for later (during planning…
harvest-session-learnings
Scan the current working session and the active task's artifacts for reusable, generalizable lessons (what was tried, what failed and why, what surprised us, what the next task should do differently) and propose anchored entries to append to the task's LEARNINGS.md, the produce-side counterpart to the ADR-0017 consume…