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 ikotelkin/claude-skills --skill working-memorygit clone --depth 1 https://github.com/ikotelkin/claude-skillsWrote 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/ikotelkin/claude-skills/working-memory)<a href="https://agentmods.dev/skills/ikotelkin/claude-skills/working-memory"><img src="https://agentmods.dev/badge/skills/ikotelkin/claude-skills/working-memory/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/ikotelkin/claude-skills/working-memory"><img src="https://agentmods.dev/badge/skills/ikotelkin/claude-skills/working-memory.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 | $0.00147 | $0.01185 |
| Opus 5.5 | $0.00059 | $0.00474 |
| Sonnet 5.5 | $0.00029 | $0.00237 |
| Haiku 4.5 | $0.00015 | $0.00119 |
Grade A, and why
working-memory 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 — 64 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Working Memory
Why
A long session runs out of context and gets compacted — the transcript is squeezed into a summary, and the details drop out: decisions, edge-cases, what was already tried and failed. Work quality falls, and the next session starts blind — the user feels "didn't we already do this?".
Root cause: the important state lived only in the conversation — volatile memory that dies on compaction and never crosses into a new session. The fix is to stage that state onto a medium that survives: a WORKING.md file in the repo.
The three levels
| Level | Medium | Lifespan |
|---|---|---|
| Short-term | session context | hours — dies on compaction |
| Mid-term | WORKING.md in the repo |
days–weeks — while a stage runs |
| Long-term | the assistant's memory (e.g. Claude Code memory files) | months — survives everything |
This skill owns the mid-term level and the bridge up to long-term. Flow: context → WORKING.md → (distill at stage end) → long-term memory. WORKING.md is a staging area for memory, not a parallel store.
The file
WORKING.md lives at the repo root (a working thread in a subfolder may have its own, e.g. docs/research/WORKING.md). Four sections:
- Current stage — one or two lines: where we are right now.
- Decision log — datestamped entries (absolute dates), newest-relevant on top: what was decided/done and why.
- Tried / didn't work — dead ends, so they aren't re-attempted.
- Next — open tails, TODOs, the obvious next step.
Copy assets/WORKING.template.md as the starting point.
Operations
Setup (once per project)
When a project has no WORKING.md and the work is non-trivial: copy the template to the repo root, fill Current stage from what's known, and add a pointer to the project's CLAUDE.md so it's read on every future session (snippet in assets/CLAUDE-snippet.md). If there is no CLAUDE.md, create one with just that pointer.
Start / resume (every session)
At the start of work on a project, read WORKING.md before acting — it restores the stage state that context loss erased. Check subfolders for thread-specific WORKING.md too. If asked to "resume", this is the anchor.
What ships with it
4 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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 · +1 lines · +147 tokens per session 334d6f244d2d
- 1mo ago First seen · 63 lines · 0 tokens per session scan A 03ece3a3be28
working-memory is a skill published in the GitHub repository ikotelkin/claude-skills (3 stars, last pushed 1mo ago), licensed MIT. It adds 147 tokens to every session and 1,185 once invoked, about $0.0006 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 2026-08-31.
Other skills, from other repositories
run-memory
Shared step of the marketing playbooks (analyze-performance, run-paid-campaigns, research-competitors, email-lifecycle): reads the last run's record, checks whether its recommendations were done and whether they worked, and saves a short record of this run for the next one. Loaded by those playbooks; not for direct…
company-brain
Your team's shared, AI-ready knowledge base — people, companies, meetings, SOPs, and decisions structured so an agent can answer on your team's behalf. Team-scope sibling to second-brain — personal notes go there, not here. Modes — capture, compile (wiki pages + INDEX.md), query (trust-weighted, saved to outputs/)…
second-brain
When you want to capture into, compile, query, lint, or connect your personal Second Brain. Wraps the Karpathy LLM Wiki schema (Obsidian or any markdown vault) — raw/ (unprocessed sources), wiki/ (AI-compiled interlinked topic pages), outputs/ (generated artifacts). Vault defaults to…
claude-reflect
Self-learning system that captures corrections during sessions and reminds users to run /reflect to update CLAUDE.md. Use when discussing learnings, corrections, or when the user mentions remembering something for future sessions.
save-learning
Shared step of the marketing playbooks (make-static-ads, analyze-performance, email-lifecycle, write-outreach): saves a short brand rule or finding in the human's own words, one idea per rule, retires the playbook's own older rules it replaces, and reads the result back to confirm. Loaded by those playbooks; not for…
learn-from-fix
Capture Elixir/Ecto/LiveView lessons and Hex API rules. Use after corrections or when asked to document learning, record a lesson, prevent a fixed mistake, or remember package guidance with --library.