| name | multi-agent-improve |
| description | Create MAPS M11 Observe/Improve for a multi-agent system. Use after M10 Runtime Readiness to observe live system signals, classify evidence, incidents, traces, eval failures, user feedback, and operator findings, and route improvements across M2-M10 and child APS phases. |
Multi-Agent Observe Improve
Versioning
Current version: 0.2.0.
Follow semantic versioning for this skill:
- Patch: wording, examples, references, or small workflow clarifications.
- Minor: new outputs, new required steps, new helper behavior, or expanded workflow capability.
- Major: renamed outputs, changed artifact contracts, removed behavior, or incompatible workflow changes.
When changing this skill, update Current version and add a Changelog entry with the date, version, and short summary of behavior changed.
Changelog
- 2026-06-24 - v0.2.0 - Expanded M11 from Improve to Observe/Improve; live observation moved here from M10.
- 2026-06-20 - v0.1.0 - Created M11 Improve for evidence-backed multi-agent system iteration.
Use /multi-agent-improve at M11 after M10 Runtime Readiness. M11 watches the live multi-agent system, captures evidence, and turns that evidence into a routed improvement backlog across MAPS system phases and child APS phases.
Project foundation updates
At the start of every project run, look for project-foundation.md. If it exists, read Persistent Memory Contract and use its configured notes, sources, memory, RAG, and sync rules as the project defaults. If .maps/foundation-preferences.json exists, use it as the structured preference source for automation.
When this skill creates durable knowledge, write it through the shared MAPS memory helper. The helper gives this skill its own named note under the configured notes root, mirrors that note into the configured RAG location when one exists, appends MAPS Skill Run Log, and records a RAG reindex manifest.
At the end of the run, call the helper after creating the primary output artifact:
python "$CODEX_HOME/skills/foundation/scripts/maps_memory.py" complete-run --project . --skill /multi-agent-improve --phase M11 --output "<primary artifact path>" --summary-file "<primary artifact path>" --memory-updates "<notes, sources, memory, or RAG updates>"
If the helper is unavailable, manually append the timestamp, skill, phase, output path, memory updates, and short note to project-foundation.md, then update this skill's named note in <notesRoot>/maps-runs/.
Research and Recommend
Read assets/maps/catalogs/multi-agent-research-sources.md when available. Use OpenTelemetry evidence, OpenAI tracing guidance, postmortem practice, backlog refinement, NIST AI RMF continuous improvement, and eval failure analysis to observe, classify, and route improvement work.
Ask exactly one question at a time. Do not present the user with a multi-question form, checklist, or table to fill out.
Start with:
What live evidence should drive this observation and improvement pass?
Then ask only the next missing question needed to define observation signals, classify findings, prioritize risk, choose phase routing, assign owners, define proof, and decide whether the next iteration restarts at M2, M3, M4, M5, M6, M7, M8, M9, M10, or a child APS phase.
Workflow
- Read M0-M10 artifacts, runtime-readiness evidence, eval results, traces, incident notes, user feedback, operator findings, and child APS improvement outputs.
- Review live observation signals: handoffs, task state, tool calls, cost, latency, quality, approvals, escalations, incidents, and user-visible outcomes.
- Classify each finding by source, severity, affected user/system behavior, phase owner, likely root cause, proof needed, and urgency.
- Route each finding to the proper phase: M2 roster, M3 contracts, M4 coordination, M5 child APS buildout, M6 capabilities, M7 orchestration, M8 experience, M9 evaluation, M10 runtime readiness, or a child APS phase.
- Split large findings into bounded improvements with owner, expected artifact, test/eval proof, release risk, and rollback or observation needs.
- Decide what stays as a backlog item, what blocks release, what requires human approval, and what is explicitly rejected.
- Update the observation log, improvement backlog, and next MAPS iteration entry point.
- Produce
m/<project-handle>/observation-improvement.md from templates/multi-agent-improve.md.
Completion report
When the skill is complete, tell the user explicitly. Do not end with only files changed or raw output.
Report:
- Completion status: complete, blocked, or needs more answers.
- Outcome: the concrete observation/improvement artifact, routed backlog, or iteration decision produced.
- Key decisions or changes made.
- Memory update: whether the shared MAPS memory helper ran, what note/run log was updated, and what RAG or notes locations need syncing.
- Next skill: the earliest MAPS or APS phase needed for the next evidence-backed iteration.
If the skill is blocked, say what answer, artifact, access, approval, or tool is needed before the next skill can run.
Output
m/<project-handle>/observation-improvement.md
- Observation signals, evidence inventory, and classification
- Routed improvement backlog
- Owner, proof, release-risk, and observation requirements
- Next MAPS/APS iteration entry point
Guardrails
- Do not treat vague preference as evidence without labeling it as opinion or taste.
- Do not route every issue to Build. Many system failures belong in roster, contracts, coordination, capabilities, orchestration, experience, evaluation, or observation.
- Do not silently expand authority or autonomy as an improvement.
- Do not close an improvement without proof or an explicit reject/defer decision.