outerloop

July 15, 2026 · View on GitHub

Outerloop

outerloop

Own the Outer Loop. Evidence → Verdict → Answerability.

Governance primitives for agentic engineering — any harness, any team size.

Companion to loop-engineering (inner loops) and harness-foundry (composable harness runtime). Works standalone with Cursor, Claude Code, or custom agents.

The loop in 60 seconds

Agent run  →  Evidence package  →  Human verdict  →  Ledger  →  Answerability
(inner)         (what happened)      (why ship/block)   (provenance)  (reconstruct why)

Humans define constraints and taste. Agents produce evidence. Humans issue verdicts with captured rationale. The system guarantees you can explain why something shipped.

→ Core concepts (5 min) · vs alternatives

Try it now

npx @cobusgreyling/outerloop init
npx @cobusgreyling/outerloop evidence package --run-id latest
npx @cobusgreyling/outerloop verdict issue \
  --evidence-id <id> --decision ship --rationale "Tests pass; scope reviewed."
npx @cobusgreyling/outerloop ledger why <id>

Or clone and run the full demo:

git clone https://github.com/cobusgreyling/outerloop.git
cd outerloop && pnpm install && pnpm build && pnpm demo

Example demo output

=== 1. Package evidence ===
  Run: 2026-07-08T10:01:59Z | Risk: 4/10
Evidence ID: 97fb7345-6849-41a1-bb68-f9446bf6824b

=== 2. Issue verdict ===
✓ Verdict recorded: ship

=== 3. Reconstruct answerability ===
# Answerability Chain: 97fb7345-...
## Verdict
- Decision: **ship**
- Rationale: Report-only daily triage: no code changes, tests pass.

→ QUICKSTART.md

Choose your harness

Persona	Get started
Any agent	`npx @cobusgreyling/outerloop init`
Cursor	`npx @cobusgreyling/outerloop init --with-cursor`
Claude Code	`npx @cobusgreyling/outerloop init --with-claude-code`
loop-engineering	`npx @cobusgreyling/outerloop init --with-loop-engineering`

outerloop evidence package --run-id latest --from cursor    # or claude-code, custom-harness
outerloop verdict review <evidence-id>
outerloop ledger why <evidence-id>

→ Adoption guide · Cursor-only example

Progressive adoption

Level	What you get	Docs
1 — Core loop	evidence, verdict, ledger	concepts
2 — Teams	taste, policy, CI gate	patterns · GitHub Action
3 — Factories	dashboard, coordination, audit	architecture

CLI essentials

Command	Purpose
`init`	Scaffold `.outerloop/`, harness, policy, taste
`evidence package`	Package agent run into EvidencePackage
`verdict review` / `issue`	Human decision with mandatory rationale
`ledger why`	Reconstruct answerability chain
`audit`	Score governance health

→ Full reference: docs/cli.md · Programmatic API: docs/api.md

CI integration

jobs:
  evidence:
    uses: cobusgreyling/outerloop/.github/workflows/evidence-gate.yml@main
    with:
      run-id: ${{ github.sha }}
      evidence-source: custom-harness

→ docs/github-action.md

Examples & patterns

Resource	What it shows
full-factory	Core evidence → verdict → ledger
cursor-only	Cursor without loop-engineering
cursor-daily-triage	Daily Cursor governance cadence
ci-gate-before-merge	CI evidence gate pattern

Status

v0.3.2 — Adoption release (CLI bin fix, contributor backlog). Phases 0–3 and v2 complete.

Phase	Status
Phase 0–1	✅ Core + evidence/verdict/ledger
Phase 2	✅ Taste + policy + integrations
Phase 3	✅ Harness, cognitive, audit, brownfield
v2	✅ Ink TUI, web dashboard, coordination

→ CHANGELOG.md · ROADMAP.md

Development

git clone https://github.com/cobusgreyling/outerloop.git
cd outerloop && pnpm install && pnpm build && pnpm test
pnpm demo

→ Contributor start here · CONTRIBUTING.md · Good first issues
Code of Conduct · Security policy

Contributors

Name	GitHub	Role
Cobus Greyling	@cobusgreyling	Creator & maintainer

→ CONTRIBUTORS.md · contributors graph

Contributing philosophy

Make answerability cheap and reconstruction trivial.
Treat taste as a versioned, first-class engineering artifact.
Separate agent capability (inner) from human agency (outer).
Build for the accountable owner, not just the implementer.

An agent can write it. Before it reaches users, someone must explain why it should exist, why it's safe enough, and what they will do when it is wrong. That is the work of the outer loop.

Built on loop-engineering, Addy Osmani's outer loop thinking, and the broader agentic engineering community.