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 sjarmak/coding-agent-workflows --skill working-set-snapshotgit clone --depth 1 https://github.com/sjarmak/coding-agent-workflowsWrote 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/sjarmak/coding-agent-workflows/working-set-snapshot)<a href="https://agentmods.dev/skills/sjarmak/coding-agent-workflows/working-set-snapshot"><img src="https://agentmods.dev/badge/skills/sjarmak/coding-agent-workflows/working-set-snapshot/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/sjarmak/coding-agent-workflows/working-set-snapshot"><img src="https://agentmods.dev/badge/skills/sjarmak/coding-agent-workflows/working-set-snapshot.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 | $0.00116 | $0.01389 |
| Opus 5.5 | $0.00046 | $0.00556 |
| Sonnet 5.5 | $0.00023 | $0.00278 |
| Haiku 4.5 | $0.00012 | $0.00139 |
Grade A, and why
working-set-snapshot 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 — 127 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Working-Set Snapshot
Compaction compresses a conversation without knowing which parts were load-bearing. A decision made an hour ago and a paragraph of tool output are equally eligible to be summarized away. The fix is a small file the agent writes and keeps current, and a mechanical hook that refuses to let compaction proceed until it exists.
The split matters. The hook does file IO, hashing, and timestamps, and makes no judgment about content. The agent makes every judgment about what goes in the file. Nothing in the script tries to infer what mattered.
This complements handoff-doc, which writes a document on purpose when you
choose to end a session. The working set covers the boundary you do not choose.
The file the agent maintains
Four sections, kept short. The full contract is in STATE-FILE-CONTRACT.md, which is worth shipping next to the script so the agent can read it.
- Doing — the task in progress, in enough detail to resume mid-step.
- Decided — decisions already made and not to be relitigated, with the reason.
- Blocked — what is stuck and on what.
- Next — the immediate next action.
Update it when any of those four change, not on a timer. A file that exists and never changes is the failure mode the hook's hash check is looking for.
What the hook does at each boundary
- If the working-set file is missing, return
continue: falsewith an instruction to write it. It does this at most--block-ceilingtimes (2 by default) and then lets compaction through. An unbounded block wedges the session. - Otherwise, copy the file to a timestamped snapshot stamped with the time, a SHA-256 of the content, and, when Git is available, the branch, the head, and the count of dirty paths.
- Warn without blocking if the file has not been touched in
--stale-minutes(90 by default), or if its content hash is identical across the last three snapshots. Warnings never block, because there is no further turn in which the agent could act on a block.
What ships with it
3 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 First seen · 127 lines · 116 tokens per session scan A d4b132beb5ea
working-set-snapshot is a skill published in the GitHub repository sjarmak/coding-agent-workflows (7 stars, last pushed 7d ago), licensed MIT. It adds 116 tokens to every session and 1,389 once invoked, about $0.0005 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 2026-09-16.
Other skills, from other repositories
digital-brain
This skill should be used when the user asks to "write a post", "check my voice", "look up contact", "prepare for meeting", "weekly review", "track goals", or mentions personal brand, content creation, network management, or voice consistency.
context-bankruptcy
Declare bankruptcy on a long-lived AI agent's accumulated memory — audit what it currently believes, separate ground truth from stale and wrong, purge deliberately, restate the truths that survive, and log what was lost. Use when an agent keeps acting on outdated facts, contradicts itself across sessions, 'remembers'…
relevance-coarse-filter
Cheap, high-recall first-pass filter that removes obvious junk from a detector candidate pool before expensive story-origin research and PR judgment. Decides keep, monitoronly, or reject — never ranks, writes angles, verifies dates, or decides whether to pitch.
context-engineering-review
Review what an LLM feature or agent actually puts in its context window — and find what's bloating, missing, or fighting itself. Use when asked to review a system prompt and context assembly, cut token usage without losing quality, debug an agent that ignores instructions, or audit how retrieval results, history, and…
memory-file-maintenance
Keep your AI memory/context file (MEMORY.md, CLAUDE.md, custom instructions) healthy over time — pruning the stale, adding the new, and keeping it sharp so your AI keeps getting you right. Use when asked review my memory file, my AI context is outdated, clean up my CLAUDE.md, or maintain my AI instructions. Produces a…
planning-with-files
Persistent file-based planning for multi-step AI-agent work. Keeps taskplan.md, findings.md, and progress.md on disk; lifecycle hooks inject selected project planning context. Automatic recovery reads project planning files only. Explicit session-catchup.py --metadata reads same-project local agent session records and…