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 VKirill/claude-lane-stack --skill winnowgit clone --depth 1 https://github.com/VKirill/claude-lane-stackWrote 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/vkirill/claude-lane-stack/winnow)<a href="https://agentmods.dev/skills/vkirill/claude-lane-stack/winnow"><img src="https://agentmods.dev/badge/skills/vkirill/claude-lane-stack/winnow/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/vkirill/claude-lane-stack/winnow"><img src="https://agentmods.dev/badge/skills/vkirill/claude-lane-stack/winnow.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 | $0.00085 | $0.00579 |
| Opus 5.5 | $0.00034 | $0.00232 |
| Sonnet 5.5 | $0.00017 | $0.00116 |
| Haiku 4.5 | $0.00009 | $0.00058 |
Grade A, and why
winnow 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 24d 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.
What it actually says
winnow
winnow sits between Claude Code's tools and your context. After Read, Bash, or Grep returns a large result, winnow splits it into blocks, asks a System One model (TypeSafe's Jev, or an LLM-backed adapter) one yes/no question per block ("is this block needed for the current task?"), and replaces the low-probability blocks with a stub.
What a stub looks like
[winnow] Lines 41-188 (148 lines) hidden: judged unlikely to matter for the current task (relevance <= 0.22).
[winnow] Summary: Boilerplate license header and the argparse setup for flags that are not part of this task.
[winnow] Full text cached as key a1b2c3d4e5f6. Call winnow_recall(key="a1b2c3d4e5f6", start=41, end=188) if you need it.
How to treat it
- The summary is written by a cheap model and preserves identifiers, paths, numbers and error text. Trust it for orientation, not for exact values.
- If the hidden lines might matter after all, call the
winnow_recalltool with the key. Addstartandendto fetch only part of the cached text. Do not re-run the original command just to see the hidden part; recall is cheaper and does not re-trigger side effects. - Outputs that show an error are never pruned. If you see a stub, the judge believed the output was clean.
- Line numbers in a stub refer to the original file or output. Claude Code renumbers the rewritten result from 1, so the numbers you see in the margin of a pruned
Readdo not match the file; use the stub's range withReadoffset/limitorwinnow_recallto reach the original lines.Editis unaffected because it matches on text, not line numbers.
Tools
winnow_recall(key, start?, end?): return the cached full text for a key.winnow_stats(): tokens saved, outputs rewritten, and the regret rate (how often a pruned key was later recalled).
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.
- 24d ago First seen · 29 lines · 85 tokens per session scan A 76a48abb7fb5
winnow is a skill published in the GitHub repository VKirill/claude-lane-stack (122 stars, last pushed today), licensed MIT. It adds 85 tokens to every session and 579 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 2026-09-23.
Other skills, from other repositories
catchup
Restore context after /clear by summarizing recent work and project state.
learning
Extracts reusable patterns from sessions. Use at session end to capture debugging insights and project-specific knowledge.
session
Session lifecycle management. Parent skill for session-related skills: learning (pattern extraction) and compact (context compression).
recall
Post-compaction context recovery. Reconstruct working context from session transcripts using MCP tools or CLI. Use when user says "recall", "recover context", "what was I doing", "after compaction", "lost context", or when context has been compacted and needs reconstruction.
vault
Context vault operations: read/update todos, add log entries, search vault content, read strategy/project docs, update project files. Use when user says "vault", "add todo", "update todos", "log entry", "search vault", "check todos", "context vault", "add to log", or any operation on a personal context vault.
session-learn
Meta-skill that analyzes past session transcripts to derive new skills, CLAUDE.md rules, and MEMORY.md entries from real usage patterns. Use when Use when user says "session learn", "derive skills", "learn from sessions", "what patterns", "analyze sessions", "extract rules", "what should I add to claude md", "improve…