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 agentmods add skills/bennyoooo/airbot/optimize-skillnpx skills add Bennyoooo/Airbot --skill optimize-skillgit clone --depth 1 https://github.com/Bennyoooo/AirbotWhat 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 | $0.00072 | $0.00876 |
| Opus 5 | $0.00036 | $0.00438 |
| Sonnet 5 | $0.00014 | $0.00175 |
| Haiku 4.5 | $0.00007 | $0.00088 |
Grade A, and why
optimize-skill 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 2d 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 — 70 lines — stays where its author put it; the contents beside it link to each section on GitHub.
optimize-skill
Make a skill measurably better without uncontrolled drift. This is an agent-in-the-loop loop, not a hands-off run: the CLI owns the deterministic machinery (scoring, edit budget, rejected-edit buffer, the gate, atomic promote/revert); you own the reasoning (running the skill, judging prose outputs, proposing edits). Expect several turns per optimization.
Preconditions
- The skill has an eval manifest (
eval.yaml) with real tasks. If it has none, stop and offer to create one (create-skill) — optimization cannot run without an eval set. - Optimization edits a managed copy, never the installed symlink target.
The loop (repeat until the gate stops improving)
-
Rollout. For each eval task
input, run the current skill yourself and collect its output. Writerollouts.json:[{ "taskId": "...", "output": "..." }]. -
Score.
scripts/optimize.sh score --eval eval.yaml --rollouts rollouts.json --skill <name> --jsonDeterministic tasks are scored for you.
agent-judgetasks come back as pending — score those yourself against each task's rubric and fold them into the aggregate. Record the current score. -
Reflect. Read the failing trajectories. Diagnose why they failed (this is your job — the CLI never judges why). Propose a small set of structured edits to
SKILL.md. Writeedits.json: an array of{ op: append|insert_after|replace|delete, target?, content?, sourceType: "failure"|"success", supportCount? }. Prefer failure-driven edits. -
Apply (bounded).
scripts/optimize.sh apply --skill <name> --skill-dir <live-dir> --edits edits.json --step <n> --total <N>The CLI caps edits at the budget (annealed over steps), skips edits to the protected
SLOW_UPDATEregion, and writes a candidate copy. Note the candidate dir it prints. -
Validate. Re-run rollout + score against the candidate (steps 1–2 pointing at the candidate dir), including the held-out tasks. Then gate:
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.
- 2d ago First seen · 70 lines · 72 tokens per session scan A 64bb2d4191be
optimize-skill is a skill published in the GitHub repository Bennyoooo/Airbot (22 stars, last pushed 4d ago), licensed MIT. It adds 72 tokens to every session and 876 once invoked, about $0.0004 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
shadow-verify
Dispatch a parallel adversarial verifier wave after any high-stakes sub-agent investigation (code reviews, audits, findings reports, large refactors, gap analyses) — or whenever a sub-agent asserts a claim with high-confidence language ("confident", "certain", "clearly", ≥80%), since confidence is a trigger, not a…
ship
Release pipeline for already-done local work. Dispatches /ground-state pre-flight, runs the project test suite, drafts a commit message, pushes, and opens a PR with a structured verification summary. Use when local changes are ready to hand off to review — e.g. 'ship this', 'push and open a PR', 'release this work'.…
ground-state
Before starting any non-trivial implementation, run a pre-flight reconnaissance pass to triangulate git state, project infrastructure, and prior-session memory. Auto-assembles a verified grounding preamble — a session-scoped artifact the orchestrator pastes verbatim into every subsequent sub-agent brief — eliminating…
refactor
Orchestrates safe, large-scale structural changes across a codebase — symbol renames, API migrations, pattern standardizations, layer restructurings. Enumerates all affected sites, groups them into dependency layers via the DAG executor, applies changes in parallel per layer with worktree isolation, and verifies…
simplify
Discovers incidental complexity, duplication, and dead code in a codebase and produces a ranked, behavior-preserving reduction plan — optionally applying safe changes. Dispatches four parallel read-only discovery lenses (clone detection, dead code, complexity hotspots, wrong abstraction), synthesizes into a…
ground-claim
Grounds capability claims with file-read evidence. Default mode answers meta-capability questions ('what does X enable') with path:line citations. Pass mode: runtime-wiring with a claims list to trace actual runtime execution paths — call sites, DI registration, middleware — and get CONFIRMED/UNVERIFIED/REFUTED…