How it works

A complete Change from demand to continuity.

1. Continuity first

retornatus wake
# or: retornatus wake --bridges

Wake reconstructs Environment capabilities, Change ids, Rules, Learnings, and rebuilds the disposable SQLite index from files.

2. Demand and Situation (requirements analysis)

A Demand is the human intent. Situation is the requirements-analysis step: enough understanding to formalize a Contract — not a full design doc.

retornatus change elicit --demand "…" --what "…" --done "…"
# When questions remain (exit 1), ask them in chat, then record:
retornatus change elicit --demand "…" --answer "scope=…" --answer "actors=…" --write situation-draft.md

Exit 0 means Situation looks sufficient; exit 1 means focused questions remain (at most five, with suggested options). The agent should ask those in chat before inventing a Contract. Repo signals (stack, tests, CI) and kickoff files (prd.md, docs/brief.md, …) feed elicitation — do not re-ask what they already answer.

3. Contract

An active Contract is the authoritative obligation: WHAT, constraints, DONE.

retornatus change create …          # activates when Situation is sufficient
retornatus change create … --draft-contract
retornatus change activate C-0001   # later
retornatus gate contract C-0001     # must exit 0

Material changes archive the prior version (change reopen) — Contracts are not silently mutated.

4. Action and Tasks

An Action is a bounded unit of work under Authority. Tasks appear when decomposition helps; declaration order is not an automatic dependency chain.

retornatus change create … --task "Implement" --task "Test" --depends "1:0"
retornatus loop next C-0001
retornatus task start C-0001/T-001

loop next never returns BLOCKED work as “next”.

5. Specialization (optional — as soon as needed)

# From a chat prompt, before Action exists:
retornatus skill need --prompt "…"
retornatus skill create --need "…"

# Or once an Action exists:
retornatus skill need --action C-0001/A-001
retornatus skill create --need "…" --action C-0001/A-001
# agent researches current docs → fills RESEARCH + PROCEDURE
retornatus gate skill-research S-0001
retornatus skill activate S-0001
retornatus skill export S-0001

Governed bypass: skill activate --force --reason "…" records Bypass + Decision.

6. Execution context

retornatus run C-0001/A-001
retornatus run C-0001/A-001 --strict-policy

Retornatus assembles Rules, Learnings, Skills, Boundaries, and Policy — it does not run the model. The host agent implements.

7. Evidence and Assurance

retornatus evidence add -c C-0001 -t test_result -s "…" \
  --source pytest --claim C-0001/claim-done-1
retornatus gate evidence C-0001
retornatus verify C-0001

Verdicts: SATISFIED · NOT_SATISFIED · INCONCLUSIVE. Evidence must bind to Claims; staleness uses path-aware commit:<sha> when available.

Independent review when needed: assurance plan / assurance review / run --assurance.

8. Problems become structure

retornatus finding add --change C-0001 --observation "…"
retornatus question open --change C-0001 --statement "…" --finding C-0001/F-001
retornatus question resolve C-0001/Q-001 --summary "…" --evidence C-0001/E-001
retornatus question reopen C-0001/Q-001   # if the condition returns

Verifiable Questions require Evidence — a summary alone is not establishment.

9. Adaptation

retornatus change learn --title "…" --body "…"
retornatus skill evolve S-0001 --note "…"
retornatus rule propose --statement "Do not …" --applicability "…"
retornatus decision record --kind APPROVE_RULE_ACTIVATION \
  --subject R-0001 --summary "…" --confirm R-0001
retornatus rule activate R-0001 --decision D-0001

Learning informs; Rules constrain. Activation always needs a Human Decision.

10. Restart

Delete .retornatus/index/retornatus.db and run wake — semantic history in files remains; the index rebuilds.

Full acceptance narrative: PRD §69.