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 Osipchuk/agent-skills --skill steelman-then-breakgit clone --depth 1 https://github.com/Osipchuk/agent-skillsWrote 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/osipchuk/agent-skills/steelman-then-break)<a href="https://agentmods.dev/skills/osipchuk/agent-skills/steelman-then-break"><img src="https://agentmods.dev/badge/skills/osipchuk/agent-skills/steelman-then-break/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/osipchuk/agent-skills/steelman-then-break"><img src="https://agentmods.dev/badge/skills/osipchuk/agent-skills/steelman-then-break.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 | $0.00194 | $0.00748 |
| Opus 5 | $0.00097 | $0.00374 |
| Sonnet 5 | $0.00039 | $0.00150 |
| Haiku 4.5 | $0.00019 | $0.00075 |
Grade A, and why
steelman-then-break 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 12d 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 — 52 lines — stays where its author put it; the contents beside it link to each section on GitHub.
steelman-then-break
Most design discussions die at "this seems good, ship it" because being the contrarian is socially expensive. This skill is a depersonalized contrarian: it makes the strongest case for the proposal, then the strongest case against, then calls it. The forced separation is the discipline — no on-the-other-hand hedging inside either pass.
Workflow
-
Restate the target in one sentence so it's unambiguous what is being evaluated. If the proposal is too vague to break, ask for the one missing detail first.
-
Pass 1 — Steelman. The strongest case for, including benefits the author may have missed. No "but", no caveats in this section. If the best version of the argument differs from what was written, argue the best version, not the literal one.
-
Pass 2 — Break. The strongest case against. Lead with what breaks first. Name the failure mode or anti-pattern explicitly. When the project has a scars file (
.claude/scars.mdat the repo root — the team's catalogued past failures), cite the relevant precedent — arguing from the team's own evidence is the point. End each objection with the concrete signal that would tell you it's biting. -
Verdict. Calibrated, not reflexively negative. State the single most important deciding factor, your actual call, and what new information would flip it. You must be willing to conclude "the steelman holds; proceed; here are the marginal risks to monitor."
Calibration
Reflexive contrarianism is the failure mode that kills this skill. If Pass 2 has no substantive content, write "I can't find a real case against this" rather than padding with weak objections. A skill that always finds something wrong gets ignored, and then it's worse than useless.
The scars file lives in the project repo at .claude/scars.md — NOT inside this
skill's folder, so the team can edit and version it without touching the
installed (checksummed) skill. It logs patterns that failed before, so the Break
pass argues from evidence instead of generic risk. references/scars.md in this
skill is the format template with a worked example: copy it to .claude/scars.md
when the team logs its first scar. No scars file just means the Break pass leans
on named anti-patterns instead.
What ships with it
1 file 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.
- 12d ago First seen · 52 lines · 194 tokens per session scan A e62e72fc6266
steelman-then-break is a skill published in the GitHub repository Osipchuk/agent-skills (5 stars, last pushed 1mo ago), licensed MIT. It adds 194 tokens to every session and 748 once invoked, about $0.0010 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 2026-08-31.
Other skills, from other repositories
audit
Audit phase. Parallel review: code quality + security + tests. Semantic dedup of cross-mode findings. Outputs PASS/WARN/FAIL per dimension. Validates spec coverage.
perf
Performance optimizer for loops, DB queries, rendering, and batch operations. Catches N+1 queries, missing indexes, and unnecessary re-renders.
_critic
HarnessX Critic (Tier 2.1) — adversarial review of evolved-skill proposals against trace evidence. Detects reward hacking and manifest/evidence contradictions. Out-of-band LLM counterpart to the in-loop deterministic critic in src/evolve/critic.rs.
simplify
Code simplification for high-complexity files. Targets deep nesting, copy-paste patterns, and god functions.
request-review
Prepare work for review so a reviewer can verify it fast and fairly. Triggers on get this reviewed, ready for review, request a review, hand off for review.
review
Review code for bugs, quality, and correctness. Triggers on review, check, audit, verify requests.