---
name: sam-agent-proof-protocol
description: Use before claiming completion, moving work to Done, posting READY FOR HERMES REVIEW, reconciling Linear/Agent Ops jobs, or reporting that a task, cron, post, send, deploy, file, API write, or automation is finished for Sam. Enforces downstream proof readback and blocker escalation.
---

# Sam Agent Proof Protocol

Use this skill whenever the work could be falsely considered done from a weak signal such as a comment, API 200, queued job, cron tick, draft, or agent self-report.

## Completion Rule

Do not claim done until the requested artifact exists in the real target system and proof was read back.

Weak signals are not completion:

- Telegram/Linear message delivered
- API returned 200
- file path was intended but not re-read
- cron lastRunStatus is ok
- post/upload queued
- email draft created when send was requested
- subagent says done
- local script exited 0 without downstream state readback

## Minimum Proof by Surface

- File/doc/code: diff, readback, syntax check, test, build, lint, screenshot, or direct inspection.
- Linear/Agent Ops: issue state, comment URL, ledger status, verified proof row, and proof artifact path agree.
- Cron/job: latest run plus expected artifact/log/output exists and is fresh.
- Browser/UI: expected visible state is present after action.
- API/database/sheets/calendar/tasks: re-query and confirm downstream record/state.
- Messaging/email/posting: sent/published/queued state matches request and is verified in the destination system.
- External publish/send: requires explicit approval unless the exact action was already approved.

## READY FOR HERMES REVIEW

Post READY only when the proof packet includes:

- exact file paths/URLs/IDs
- what changed
- verification commands/readbacks
- what was not done because it is approval-gated
- rollback instructions for file/config changes
- remaining blockers, if any

## BLOCKED

When blocked, include:

- exact missing input/approval/access/credential
- what was tried
- why it failed or why action is unsafe
- what is still executable
- best next action
- what Sam needs to decide or provide

## Linear and Ledger Reconciliation

Before counting Linear work as complete, verify all are aligned:

- Linear issue has READY/APPROVED/BLOCKED comment with proof.
- Agent Ops job is done, blocked, failed_terminal, or needs_continuation with a real next action.
- Proof row is verified when the ledger says done.
- No active resource lock remains unless work is actively running.
- Next-job selector skips approval-gated or Hermes-review-waiting work.

## Safety

Do not expose secrets. Do not mutate scheduler/config/gateway/auth/session state, publish externally, delete, rotate, or reset systems without explicit approval.