Skip to content

concepts.md

Latest commit

 

History

History
75 lines (54 loc) · 2.68 KB

concepts.md

File metadata and controls

75 lines (54 loc) · 2.68 KB
Raw

Core concepts (5-minute read)

outerloop governs the human side of agentic software work. Agents run the inner loop; humans own constraints, decisions, and explainability.

The loop

Agent run  →  Evidence  →  Verdict  →  Ledger  →  Answerability
(inner)        (facts)      (decision)   (record)    (explain why)
Concept One-line definition
Evidence Structured package of what an agent run did — plan, diffs, tests, risk.
Verdict Human production decision (ship, block, narrow, …) with mandatory rationale.
Ledger Append-only provenance chain linking evidence to verdicts and commits.
Answerability Guarantee that ledger why reconstructs why something shipped.

Progressive adoption

Level 1 — Everyone

Start here. Three commands, no framework lock-in:

npx @cobusgreyling/outerloop init
outerloop evidence package --run-id latest
outerloop verdict review <evidence-id>
outerloop ledger why <evidence-id>

Level 2 — Teams

Add organizational constraints:

  • Taste — Versioned qualitative rules (outerloop taste capture, taste apply).
  • Policy — Backpressure and sampling (outerloop policy set, policy evaluate).
  • CI gate — Reusable GitHub Action packages evidence on every run.

Level 3 — Factories

Multiple agent loops, dashboards, health scoring:

  • Harness — Declarative inner/outer boundary.
  • Coordination — Multi-loop registry and collision checks.
  • Dashboard / audit — Team visibility and governance health.

Integrations (pick your harness)

Harness Setup
Any agent outerloop init + manual evidence package
Cursor outerloop init --with-cursor
Claude Code outerloop init --with-claude-code
loop-engineering outerloop init --with-loop-engineering

Evidence source field accepts: custom-harness, cursor, claude-code, loop-engineering, manual.

Glossary (extended)

Term Meaning
Inner loop Agent capability: investigate, implement, verify, repeat.
Outer loop Human agency: constraints, sampling, verdict, ownership.
Backpressure Policy that throttles agent autonomy so humans can keep up.
Taste Qualitative judgment encoded as versioned rules.
Cognitive debt Erosion of understanding about what shipped and why.
Harness boundary Explicit split of what agents may do vs what humans must decide.

Further reading