How nullurl/marvin-adaptive-delivery works
Marvin Adaptive Delivery is an agent-facing delivery workflow and local execution control plane, not a project-management service. The principal path starts with a user goal: the Skill selects risk-appropriate governance, organizes work and evidence, and can use the A2A server to track delegated attempts and recovery before verification and acceptance. Separate capabilities initialize project governance, route model profiles, validate visual evidence, and preserve context and reusable experience. The A2A server owns execution state but leaves worker execution to its caller. Relationships for unsampled subsystems rely on README statements; source excerpts do not establish additional runtime call chains.
Main entry points: a2a_server.py A2A control plane · SKILL.md Risk-tiered flow · long-task-governance.md Agent coordination · init_project.py Project initializer
Components 19
Delivery workflow
- Delivery Skill
SKILL.md - Risk-tiered flow
SKILL.md - Agent coordination
references/long-task-governance.md - Context governance
references/context-governance.md - Experience methods
references/experience-catalog.md
Execution control
- A2A control plane
scripts/a2a_server.py - Execution database
scripts/a2a_server.py - Worker agents
- Verifier
- Recovery worker
- Model profile router
scripts/model_routing.py
Project governance
- Project initializer
scripts/init_project.py - Project templates
assets/templates - Governance checks
scripts/lint_governance.py
Evidence and validation
- Verification receipts
scripts/verification_guard.py - Visual evidence guard
scripts/visual_evidence_guard.py - Visual reference governance
references/visual-reference-governance.md - Harness and release
references/harness-and-release.md
Other components
- User
Connections 19
- User → Delivery Skill: requests delivery
README.md - Delivery Skill → Risk-tiered flow: selects flow
README.md - Risk-tiered flow → Agent coordination: coordinates work
README.md - Risk-tiered flow → Context governance: governs context
README.md - Risk-tiered flow → Experience methods: captures methods
README.md - User → Project initializer: initializes project
README.md - Project initializer → Project templates: copies templates
README.md - User → Governance checks: checks governance
README.md - Agent coordination → A2A control plane: creates tasks
README.md - A2A control plane → Execution database: persists state
README.md - A2A control plane → Worker agents: dispatches attempts
README.md - Worker agents → A2A control plane: sends events
README.md - A2A control plane → Verifier: provides candidates
README.md - Recovery worker → A2A control plane: resumes checkpoints
README.md - A2A control plane → Model profile router: resolves profiles
scripts/a2a_server.py - User → Verification receipts: runs verification
README.md - Verification receipts → Harness and release: records receipts
README.md - User → Visual evidence guard: checks receipt
README.md - Visual reference governance → Visual evidence guard: defines evidence
README.md