Sequence · 06 of 08 · One sentence

The parachute sentence, end to end

The agent never edits a file. It describes what it wants, reviews a plan with the assumptions stated, and applies it. Clearance opens PRs, humans merge the flight ones, and verify is a check against the running system, not a green pipeline.

The parachute sentence, end to end Sequence diagram of an agent handling the request to provision a Python application named parachute: the agent asks for capabilities, sends an app spec to plan, Clearance calls the planner and returns a change set with assumptions, the agent applies it, Clearance opens a tenants PR that a human merges, waits for the repos, opens the app and gitops PRs, then returns a verified result. AIRFRAME.CAPABILITIESSTACKS · SCHEMA · REGISTRYPLAN(APPSPEC)COMPILE(SPEC, REGISTRY)CHANGE SET · 5 STEPSPLAN + ASSUMPTIONS + WARNINGSAPPLY(CHANGE SET)PR 1: TENANTS (APP + 2 ENVS)MERGEREPOS READY · CICDONBOARDEDPR 2 APP REPO · PR 3 GITOPSVERIFY: 4 ENVS CONFIGUREDClaude Codethe user's sessionClearanceairframe.* toolsPlannerpure functionGitHub + Argorepos · syncHumanmerges flight PRsLEGENDCallReturnVerifiedGoverned gate

Principle honored

Plan before apply, and ambiguity is a question, not a guess: with two upper clusters and no choice, plan returns a question instead of picking one.

What changes

  • plan is idempotent: the same spec gives the same change set, and a converged app is a no-op.
  • State lives on the task_id, PR labels and the audit log, so a session can end and resume.
  • Human gates: the tenants PR and the gitops PR. Ground changes auto-merge.

What is built

Built and tested: the planner and 26 tests, including the generated XRs against the real XRD schemas, cicd.yaml against Glidepath's schema, and a real helm template. Not built: the adapters that open the PRs.

Hangar · Autopilot · plan