morpheus-ai AGENTS.md

AGENTS.md

Morpheus Bootstrap

Fetch the Morpheus manifest before making changes:

• Connect manifest: start the API/UI, then fetch /agent/connect?project_root=<PROJECT_ROOT>.
• One-command prepare: morpheus prepare-agent.
• Local handoff bundle: morpheus handoff.
• Local CLI manifest: morpheus agent-connect --json.
• Local diagnostics: morpheus diagnostics --json.
• Read WAKE.md before edits.
• Run compile and verify after meaningful changes.
• If the API/UI are unavailable, start them with morpheus serve --ui --host 127.0.0.1 --port 8000 --ui-port 5173.
• Use 0.0.0.0 only for explicit user-approved trusted LAN testing or authenticated proxy environments.

Agent sequence:

1. Fetch /agent/connect for this project root.
2. Initialize only when state.initialized is false.
3. Compile, then read WAKE.md.
4. Make the requested project change.
5. Compile again and run morpheus verify --all.

Morpheus Roadmap Autopilot Rule

When asked to continue Morpheus roadmap work:

• Do one slice only.
• Prefer tests first.
• Treat truth-layer work as the data-quality gate, not as a replacement for weights-as-memory.
• Treat adapter weights as the final memory artifact once dataset quality, eval, activation, and rollback gates exist.
• Use this roadmap unless the user explicitly changes it:
  • v0.3: semantic classifier as product core.
  • v0.4: dataset quality dashboard.
  • v0.5: adapter memory benchmark.
  • v0.6: agent memory routing.
  • v0.7: team learning loop.
• No accepted source span means no training example.
• No eval pass means no adapter activation.
• No rollback means no production activation.
• Keep morpheus wake . deterministic by default.
• Keep cloud providers opt-in.
• Never make semantic candidates active without review.
• Never tag, release, publish, or push unless explicitly instructed.
• Always run:
  • ruff check .
  • pytest tests/ -q
  • morpheus wake . --private
  • morpheus verify --all
• If touching check, also test file input, stdin input, JSON output, and exit codes.