Safe Receipt Workflow

Goal

Enable high-confidence execution for a narrow workflow while keeping a compact audit trail of what was allowed, blocked, and unverified.

Inputs

• target path
• target workflow/task
• expected output target
• security risk scope from policy/agent-safety-policy.json

Process

1. Load policy/agent-safety-policy.json and copy risk_scope into working memory.
2. Announce restricted paths and disallowed command patterns before changing files.
3. Preflight planned edits:

• if a planned edit hits a restricted path, stop and ask for explicit approval.
• if a planned action is not in allowed_tools, stop and ask for approval.

1. Execute only in-scope operations and keep command/use evidence.
2. Run project checks relevant to the request.
3. Validate no secret patterns or restricted paths were touched without permission.
4. Emit templates/completion-receipt.md with claims and evidence.
5. If the workflow needs CI, audit, or MCP-style downstream processing, also fill templates/audit-receipt.json.

Stop Conditions

• Any plan includes rm -rf, remote credentials, or blocked outbound destinations.
• Planned edits target .env, *.pem, *.key, secrets/, keys/, dist/, build/.
• No explicit policy file is available.

Output

Use templates/completion-receipt.md at end. Add templates/audit-receipt.json when a machine-readable receipt is useful.

Verification

Before finishing, check verification.md.