PIP evidence discipline

Use when an agent, operator, or worker has failed twice, is repeating the same approach, is about to hand the task back without evidence, or has claimed completion without running the relevant check.

Source and scope

Adapted from the English PIP edition of tanweai/pua:

• Source: https://github.com/tanweai/pua
• Audited commit: f16769945d42bdb568fc956d53f2c9f347d66ae2
• Audited source file: codex/pua-en/SKILL.md
• Audited SHA-256: 9ac522d2f80655853a4445a5ef0824ffdc139d737030494beb16ccbc1104f98a

This Storm integration imports the operating discipline only. It does not include plugin hooks, prompt routers, telemetry, feedback upload, leaderboard, survey, remote prompt loading, or always-on behavior.

Public boundary

This skill is for public ecdsa.fail coordination. Do not request or expose:

• private endpoints,
• account details,
• keys or tokens,
• raw logs,
• unreleased nonces,
• live candidate diffs,
• private mailbox history,
• submit authority.

When a next step needs private state, stop at a handoff boundary and name the public evidence that justifies the request.

Trigger

Activate this card when any of these are true:

• the same route, prompt, or script failed twice,
• the agent keeps changing parameters without changing the idea,
• a route is blamed on environment or tooling before verification,
• a worker asks for human action before checking public/local evidence,
• a completion claim lacks command output, validator output, or source citation,
• a compute request lacks owner, validator, kill gate, budget, or stop condition.

Do not activate for a normal first pass when a known fix is already being executed.

Protocol

1. Stop the current loop. List the last attempts and the shared assumption.
2. Read the failure signal word by word: error, rejection, empty result, stale route, missing evidence label, or user critique.
3. Refresh public truth before relying on memory: current target, public docs, fixture data, route packet, or source file.
4. Read the raw material, not a summary. For code, inspect local context around the failing line. For routes, inspect the packet, validator, and kill gate.
5. Verify assumptions with tools: path, version, input shape, validator command, budget, public source, and stop condition.
6. Invert the main assumption. If the route is assumed alive, try to kill it. If a bug is assumed local, check the caller, fixture, and validator.
7. Produce three materially different hypotheses, not three parameter tweaks.
8. Execute the smallest bounded check that creates new evidence.
9. Close the loop: record what passed, what failed, what was killed, and what must not be retried.

Output

Return this block before the next major action:

PIP discipline:
- Failure mode:
- Repeated assumption:
- Evidence read:
- Verified facts:
- Three different hypotheses:
- Next bounded check:
- Kill gate:
- Completion evidence required:

Completion gate

Do not say a route, patch, dashboard, or note is complete until the relevant check has been run or the missing check is named as a blocker.

For this repo, useful completion evidence includes:

• scripts/redaction-check.sh
• scripts/check-public-harness.sh
• fixture dashboard preview using demo/public data only
• route packet fields filled with owner, validator, budget, and stop condition
• public-note language that avoids win claims unless the submit gate is met

Exit report

If the route still cannot advance after the protocol, return a structured handoff instead of a vague failure:

• verified facts,
• excluded possibilities,
• smallest remaining uncertainty,
• next public check,
• exact private boundary, if any,
• reason to proceed, park, kill, or request human review.