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 ashutoshsinghpr7/wikiskill --skill wikiskill-evolvegit clone --depth 1 https://github.com/ashutoshsinghpr7/wikiskillWrote 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/ashutoshsinghpr7/wikiskill/wikiskill-evolve)<a href="https://agentmods.dev/skills/ashutoshsinghpr7/wikiskill/wikiskill-evolve"><img src="https://agentmods.dev/badge/skills/ashutoshsinghpr7/wikiskill/wikiskill-evolve/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/ashutoshsinghpr7/wikiskill/wikiskill-evolve"><img src="https://agentmods.dev/badge/skills/ashutoshsinghpr7/wikiskill/wikiskill-evolve.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00048 | $0.00666 |
| Opus 5.5 | $0.00019 | $0.00266 |
| Sonnet 5.5 | $0.00010 | $0.00133 |
| Haiku 4.5 | $0.00005 | $0.00067 |
Grade A, and why
wikiskill-evolve 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 — 69 lines — stays where its author put it; the contents beside it link to each section on GitHub.
WikiSkill Evolve
Run Google's WikiSkill evolution loop (arXiv:2608.27454) — a faithful
implementation with Hermes as the reference backend. The agent evolves its own
skills: raw sessions → a maintainer distills failure patterns into a
persistent wiki → a proposer writes a candidate skill → a validation gate
accepts it only if R_val > R_best (git rollback otherwise).
When to use
- You want an agent's own experience (traces) turned into reusable skills
- You want to test whether a candidate skill actually helps, with statistics instead of vibes
- You're running the paper's protocol on your own tasks
Install
pip install wikiskill # Python ≥ 3.10; works with Hermes, Claude Code
Run the loop
wikiskill init myws # workspace + auto-graded bench
wikiskill evolve myws --iters 3 # train → maintain → propose → gate
wikiskill status myws # baseline, r_best, skill state
wikiskill compare wsA wsB --iters 5 # paired exact-binomial comparison
wikiskill transfer src dst # copy accepted skills to another ws
Reading the output
runs/state.json—baseline(S₀ on val),r_best,next_iterwiki/log.md— every maintenance/proposal/gate decision with evidencewiki/patterns/— distilled failure patterns (the raw material)wiki/skill-impact.md— rejected proposals stay visible (paper requirement)- Gate verdicts:
ACCEPTED(R_val > R_best, git commit),REJECTED(rolled back),no_action(proposer declined — a valid outcome)
Backends
wikiskill init myws --backend claude # Claude Code as the worker
Hermes is the reference backend; Claude Code ships in the box; codex/opencode are on the roadmap. All speak open SKILL.md, so evolved skills transfer.
Honest-expectation notes
- Each iteration costs ~$0.09 on free-tier models (gemini-lite class) —
turn budgets and
--max-turnsbound the spend - A weak model may produce
no_actioniterations — that's the gate working, not a failure; skill accumulation needs a reasonably strong proposer - The gate has rejected harmful skills in live runs — a rejection is a win
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 · 69 lines · 48 tokens per session scan A 521054c5993d
wikiskill-evolve is a skill published in the GitHub repository ashutoshsinghpr7/wikiskill (281 stars, last pushed 1mo ago), licensed MIT. It adds 48 tokens to every session and 666 once invoked, about $0.0002 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 2026-09-05.
Other skills, from other repositories
mnemosyne
Persistent cross-session memory via Mnemosyne — store, recall, and consolidate facts, preferences, and context.
unified-code-review
Risk-first code review for PRs and branch audits: blast-radius triage, agent-authored discipline (tests first, intent evidence), call-graph pincer for integration defects between modules, then structural code-judo bar. Use when reviewing PRs, auditing agent-written diffs, catching rubber-stamp green CI, or wiring bugs…
email-inbox-triage
Triage an inbox: prioritize threads, draft replies safely.
harness-optimizer
Optimize an AI agent's harness for MCP-Atlas benchmark. Use when analyzing execution traces, diagnosing failures, and proposing improved prompts, skills, or harness code.
gui-click-over-keyboard
When to prefer GUI mouse clicks over keyboard shortcuts — especially for formatting, multi-step visual tasks, and cross-application workflows.
debug-and-fix
Strategies for debugging build failures, runtime errors, writing complete output files (ICS/JSON/config), and constraint satisfaction tasks.