Agoragentic Premortem Golden Loop

npm version CI license node

Find the launch failure before your users do.

Premortem Golden Loop is a free local audit and repair-guidance CLI for plans, launches, products, and installable AI-agent repositories. It frames the launch as already failed, finds evidence-backed risks, checks the no-spend release loop, proposes bounded fixes, and rechecks what was actually resolved.

npx agoragentic-premortem-golden-loop@latest audit --repo .

The default run is local and writes inspectable artifacts. It does not upload repository contents, call a paid service, deploy, publish, transfer funds, mutate wallets, rotate secrets, or overwrite application source.

Run the audit · Review safe fixes · See sample outputs · Integrate it

The loop

Assume the launch failed
        ↓
collect likely failure reasons
        ↓
inspect repository evidence
        ↓
rank blockers and hidden assumptions
        ↓
produce repair guidance and an IDE handoff
        ↓
apply only reviewed bounded changes
        ↓
rerun and record resolved vs. still open

The tool produces evidence and recommendations. It does not certify the repository, guarantee launch success, or replace an independent security, legal, financial, or operational review.

Run the audit

Explain the boundary first

npx agoragentic-premortem-golden-loop@latest doctor --repo .

doctor reports what the tool plans to read, which local artifacts it may write, and which actions remain outside its authority.

Run the default local audit

npx agoragentic-premortem-golden-loop@latest audit --repo .

The audit combines:

• repository release premortem;
• install/config/discovery/proof/approval/test readiness;
• local no-spend Golden Loop checks;
• launch-gate findings;
• a bounded self-heal plan;
• an IDE-agent repair prompt;
• a closure loop for later reruns;
• a clearly labeled local receipt.

Add the business context

A useful premortem needs to know what is launching, who it is for, and what success means:

npx agoragentic-premortem-golden-loop@latest audit --repo . \
  --plan "Launch an OSS AI-agent audit CLI" \
  --audience "Builders preparing public agent repositories" \
  --success "A builder runs it and fixes one material launch risk"

When required context is missing, the tool asks for the next necessary input instead of fabricating a generic business conclusion.

What success looks like

Artifacts are written under:

.agoragentic/premortem-golden-loop/
├── doctor.json
├── doctor.md
├── audit.json
├── audit-summary.md
├── audit-guide.html
├── premortem.json
├── premortem.md
├── premortem-report-<timestamp>.html
├── golden-loop.json
├── golden-loop.md
├── local-receipt.json
├── healing-plan.json
├── healing-plan.md
├── ide-fix-prompt.md
├── agent-handoff.md
├── closure-loop.json
└── closure-loop.md

A green-looking report is not the goal. The useful result is a truthful separation between:

• verified resolved issues;
• remaining blockers;
• assumptions that still need owner evidence;
• recommendations that were not applied;
• checks that were not run.

Review and repair

Generate the repair plan only

npx agoragentic-premortem-golden-loop@latest heal --repo .

This writes a proposed plan without editing the repository.

Apply reviewed additive scaffolds

npx agoragentic-premortem-golden-loop@latest audit --repo . --apply-safe-fixes

The bounded apply mode may create only missing additive scaffolds such as:

• goals/workflow/safety-boundary documentation;
• agent.json;
• .env.example;
• a Golden Loop GitHub Actions workflow.

It does not overwrite existing files, edit application source, delete files, rotate credentials, install dependencies, call paid execution, deploy, publish, sign a wallet, transfer funds, or open production runtime.

Secret findings, license choices, production changes, and existing-file edits remain owner actions.

Recheck the closure loop

npx agoragentic-premortem-golden-loop@latest audit --repo .

The rerun compares current evidence with the previous local audit and records what is now verified resolved and what remains open.

For a CI gate:

npx agoragentic-premortem-golden-loop@latest run --repo . --ci --skip-network

Optional declared tests run only when you explicitly add the relevant test flag and have reviewed the repository's test contract.

What it checks

The static and local readiness pass looks for evidence such as:

• a reproducible install contract;
• an OSS license and repository metadata;
• declared test commands;
• machine discovery such as agent.json, an agent card, SKILL.md, MCP, OpenAPI, or another manifest;
• secret-shaped committed values without printing the secret;
• .env.example or equivalent configuration guidance;
• owner approval and spend boundaries;
• receipt, trace, invocation, reconciliation, or audit-proof contracts;
• health, readiness, rollback, or runbook guidance;
• alignment with Agent OS, Micro ECF, or execute(task, input, constraints) when those systems are claimed.

A missing artifact is not automatically a bug for every project. Findings should be interpreted against the declared product and launch goal.

The no-spend Golden Loop

The local Golden Loop checks readiness—not settlement:

1. install contract exists;
2. configuration and secret boundaries are explicit;
3. agent discovery exists when applicable;
4. material premortem blockers are surfaced;
5. a proof/receipt contract exists when claimed;
6. owner approval and spend authority are explicit;
7. optional public no-spend canaries respond, only when enabled;
8. an optional local target runtime responds, only when supplied;
9. declared tests pass, only when explicitly enabled.

Public canaries are off by default. When enabled, they send no repository contents and use only public no-spend Agoragentic surfaces.

npx agoragentic-premortem-golden-loop@latest run \
  --repo . \
  --allow-network-canaries

A successful local Golden Loop receipt is not a marketplace verification, settlement receipt, certification, or proof of every external outcome.

Plan-only premortem session

Run a standalone six-month failure-frame session for a strategy, hire, launch, or product decision:

npx agoragentic-premortem-golden-loop@latest session \
  --plan "Launch the product" \
  --audience "Target users" \
  --success "Observable success condition"

The session records:

• raw failure reasons;
• one investigator pass per reason;
• most likely and most dangerous failures;
• hidden assumptions;
• a revised plan;
• a pre-launch checklist;
• the evidence and unknowns behind the synthesis.

The method prompt is documented in PROMPT.md.

Integrate it

MCP

npx --yes agoragentic-premortem-golden-loop@latest mcp

Local/private HTTP agent

npx agoragentic-premortem-golden-loop@latest serve \
  --repo . \
  --host 127.0.0.1 \
  --port 8787

The server binds to loopback by default. Non-loopback binding requires an explicit token. Remote safe fixes, network probes, and test execution require separate owner-controlled flags.

GitHub Actions, Docker, and IDE agents

Use the maintained templates in templates/ and the exact setup guides in:

• Integration recipes
• External HTTP agent
• Release checklist

Templates exist for GitHub Actions, Docker/home-server operation, systemd, Codex, Claude Code, Cursor, Cline, Windsurf, and other supported IDE-agent paths.

Sample evidence

• Audit summary
• Closure loop
• IDE repair prompt
• Local receipt

Sanitized examples illustrate the output contract. They are not current audit results for your repository.

Safety boundary

By default, the tool:

• runs locally;
• makes no network calls;
• sends no repository contents, prompts, plans, reports, or receipts anywhere;
• performs no paid execution;
• makes no production mutation;
• does not automatically apply source-code changes.

Even with --apply-safe-fixes, it remains restricted to reviewed missing additive scaffolds. It never treats a model recommendation as owner authorization.

Development

Requires Node.js 18 or newer.

git clone https://github.com/rhein1/agoragentic-premortem-golden-loop.git
cd agoragentic-premortem-golden-loop
npm install
npm run check
npm test
npm run release:check

release:check runs syntax checks, local tests, a no-spend repository run, and an npm package dry run.

Where this fits

• Before launch: Premortem Golden Loop finds release risks and prepares bounded repair guidance.
• During engineering: Fable-5 performs evidence-first audits, reviews, fact checks, and repo sweeps.
• For context: Micro ECF and ECF Core govern local source context.
• For actions and receipts: Harness Core governs tool/action lifecycles.
• For hosted operation: Triptych OS runs governed deployed agents.
• For agent commerce: Marketplace and Interchange connect buyers, sellers, and markets.

Use the canonical ecosystem profile for current portfolio metadata. This README intentionally does not duplicate mutable integration counts.

License

MIT. See LICENSE.