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 Fergana-Labs/stash --skill cleanupgit clone --depth 1 https://github.com/Fergana-Labs/stashWrote 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/fergana-labs/stash/cleanup)<a href="https://agentmods.dev/skills/fergana-labs/stash/cleanup"><img src="https://agentmods.dev/badge/skills/fergana-labs/stash/cleanup.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector pass
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 | $0.00024 | $0.00552 |
| Opus 5 | $0.00012 | $0.00276 |
| Sonnet 5 | $0.00005 | $0.00110 |
| Haiku 4.5 | $0.00002 | $0.00055 |
Grade A, and why
cleanup 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 8d 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 — 55 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Cleaning up a saved library
Propose, never dispose. You do not delete, merge, retag or archive anything in this skill. You produce a list the user can act on, and you act only on items they name afterwards. An agent that quietly tidies someone's library is an agent they stop trusting.
The three piles
The Bookmarks app computes these; read them rather than deriving them.
Duplicates — the same page saved more than once, matched on a normalized
URL (protocol, www., trailing slash and tracking params ignored). The
oldest save is treated as the original.
For each group, recommend which to keep, and say why: prefer the one whose saved copy actually captured content over a link-only save, then the one with topics already on it, then the oldest. Note when the "duplicates" are not really duplicates — a URL that serves different content over time, for instance — and leave those alone.
Broken — a link the checker got a 404 or 410 from. Note it only trusts those two codes, so a broken item is genuinely gone rather than merely unreachable. For each, check whether the saved copy still has the content: if it does, the recommendation is "keep, the archive is the point now"; if it is link-only, there is nothing left and it can go.
Untagged — no topics. Usually means enrichment failed or the page had
nothing to read. Propose topics from the existing vocabulary — read it off
the Topics column's options — rather than inventing new ones; a cleanup
pass that adds twelve new near-synonyms has made things worse.
Write it
Group by pile. Each item: what it is, the recommendation, one clause of why, and a link. Lead with a one-line count of each pile so the user can decide where to spend their attention.
End with the single sentence they need: "Say which of these to action and I'll do it." Then stop.
Rules
- No action without an explicit instruction naming what to act on.
- If a pile is empty, say so in one line rather than omitting it — knowing there are no broken links is worth something.
- Never recommend deleting the last remaining copy of anything.
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.
- 8d ago First seen · 55 lines · 24 tokens per session scan A a8b0ca861e8b
cleanup is a skill published in the GitHub repository Fergana-Labs/stash (332 stars, last pushed 6d ago), licensed MIT. It adds 24 tokens to every session and 552 once invoked, about $0.0001 per session on Opus 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
antigravity-cli
Delegates large multi-step work to Antigravity CLI like a subagent—include full goal, prior decisions, scope, and constraints in prompts so isolated sessions stay aligned with the main thread (context handoff per Cognition-style delegation). Use for heavy edits, exploration, web search, and read-only plan-mode…
memoized-command-source-aggregation
Load independent command sources in parallel, then memoize the merged registry by cwd so repeated lookups stay cheap and ordered.
slash-command-skill-classification-filter
Derive a slash-command skill surface from a broader prompt-command registry by requiring descriptive metadata, excluding builtins, and admitting only a controlled set of origins plus explicit user-only commands.
lazy-heavy-command-shims
Stub heavy commands so the CLI stays light, loading the real implementation only when the command is executed.
skip-skill-discovery-on-skill-expansion
Bypass skill discovery when expanded SKILL.md or generated prompt content is being reprocessed for attachments, so non-user text does not trigger recursive discovery work.
skip-skill-listing-without-skill-tool
Omit skill-listing attachments for agents that do not expose the Skill tool, so discoverability context is only sent where it can be acted on.