skills.sh
AvailableUniversal skill installation, including Claude Code and Codex.
npx skills add https://github.com/ulpi-io/skills-autonomous-engineering --skill adversarial-verifyadversarial-verify
Makes independent skeptics try to disprove a claim and keeps it only when it survives a fail-closed vote.
WORKFLOW / 05/adversarial-verify <the claim/finding to verify, or a path to a findings list>
- 01
Rewrite the claim as a concrete, falsifiable proposition.
- 02
Choose an odd panel sized to the stakes and assign distinct failure lenses.
- 03
Give independent skeptics the ground truth and tell each to refute.
- 04
Tally the evidence with majority-survive and fail-closed quorum rules.
- 05
Return survivors and the complete rejection ledger.
▸ operating envelope
Use it deliberately.
The shortest useful definition of where this skill helps—and where it should stay out of the way.
Use when
- An agent assertion could drive an expensive or hard-to-reverse action.
- A defect, fix, benchmark win, or clean verdict needs proof from the underlying evidence.
- Several failure lenses need independent scrutiny.
Do not use when
- A fact is already machine-proven by a successful command or test.
- The verifiers cannot access the actual code, diff, reproduction, or invariant.
- The originating claimant would be its own verifier.
▸ output contract
What must be true when it stops.
Produces
- Panel size, lenses, and evidence per claim
- Survived-claim list
- Rejected-claim and counterexample ledger
- Fail-closed clean or not-clean verdict
Guarantees
- Verifier prompts are adversarial, not confirmation-seeking.
- The claimant never sits on its own panel.
- Every verifier receives ground truth rather than a summary alone.
- Ties, abstentions, dead verifiers, and insufficient evidence do not pass.
- Concrete counterevidence can outweigh bare majority assertions.
Honest failure states
- rejected — a majority refuted the claim or could not establish survival
- no_quorum — verifiers died, timed out, or returned empty evidence
- ambiguous — evidence cannot support a safe survived verdict
▸ enforcement
Know which claims are executable.
Scripts are linked when this skill owns deterministic machinery. The remaining rules are explicit operating contracts and must not be marketed as hooks.
▸ install and invoke
One skill, three honest surfaces.
Claude Code plugin
AvailableInstall the full plugin, then invoke /adversarial-verify or let routing select it from context.
Codex plugin
Work in progressA work in progress on the codex-native-plugin branch — build it, start a new session, then invoke $autonomous-engineering:adversarial-verify.
▸ composes with