AI Agent Workflow Collection

Use for OZM-governed agent-native, multi-agent, memory, MCP/tool, control-plane, operator-shell, or agent framework work that needs runtime-real evidence, user-agent parity, tool contracts, state transitions, memory boundaries, and claim ceilings.

Use for OZM-managed API contract, endpoint, request/response, schema, negative behavior, compatibility, or contract test work.

Use for OZM-governed capability evolution, evo/self-evolving agent work, self-improving coding-agent loops, capability-evolver donors, skill evolution, mutation/promotion/rollback gates, benchmark-first improvement, LLM evaluator API posture, and safe promotion of learned agent behavior.

Use whenever OZM-governed work is about to make a positive claim and the exact ceiling must be stated honestly as planned, dispatch-frozen, artifact-present, pending-controller-gate, historical-only, verified, or accepted.

Use when an OZM-governed phase, lane, milestone, release, or user-requested reference retrospective needs controlled closeout, handoff, truth alignment, full-gate trigger reconciliation, inherited-vs-fresh labeling, and unresolved debt carry-forward.

Use during admitted OZM-governed implementation to keep code decoupled, completion-directed, readable, map-aware, bounded to the frozen write-set, and efficient under scoped packet gates without weakening final proof.

Use for OZM-governed context engineering: compression, context degradation, context budgeting, progressive disclosure, large-output routing, filesystem-backed context, post-compaction recovery, and long-horizon memory/working-index health.

Use for OZM-managed schema migration, data migration, rollback, seed, index, storage table, enum, or persistence change work.

Use for OZM-managed deployment, release, rollback, environment promotion, infrastructure command, CI/CD, or runtime rollout work.

Use before starting, replaying, or replacing an OZM-governed writer lane to freeze the reference row, write-set, controller-owned surfaces, admission order, current claim ceiling, change class, gate tier, and runtime carrier posture.

Use when OZM-governed work creates, rewrites, deepens, audits, or iteratively improves a text artifact such as a plan, spec, report, analysis, handoff, research note, prompt package, roadmap, design doc, or acceptance narrative.

Use when OZM-governed work needs bug reproduction, root-cause isolation, minimal repair, and debug recording without widening scope or confusing harness noise with product truth.

Use for OZM-managed expert review gates after governance is frozen, without loading archived standalone reviewer skills.

Use when an OZM-governed work packet depends on secrets, providers, browsers, deployments, project environment entrypoints, orchestrators, local runtimes, or remote services and must not be admitted until the prerequisite contract and reachability posture are proven.

Craft and audit GPT Image 2 / image-2 prompts and visual briefs under OZM governance. Use for prompt-only planning, image-2 brief refinement, and project asset prompt evidence posture; actual generation remains owned by imagegen.

Use for OZM new-project pre-start setup or audit: TruthDocs surfaces, task/goal loops, Goal/runtime reentry gates, loop output discipline, placement/environment gates, smoke/probe script placement, and bounded run-until-done autonomy.

Use for OZM-managed logging, metrics, traces, alerts, SLO, error reporting, or runtime diagnosis work.

Use for OZM-managed performance, load, latency, memory, throughput, profiling, regression, or budget work.

Use when OZM-governed control surfaces such as task cards, receipts, ledgers, state files, evidence hashes, audit receipts, active windows, and maps must stay synchronized with live progress instead of drifting into stale summaries or proof-chain noise.

Use when OZM-governed work touches known repeated failure modes such as stale summaries, ownership drift, self-promoted evidence, nonstart loops, late prerequisite discovery, evidence hash cascades, gate-noise loops, runtime-carrier mismatch, masking, or false blockers and needs extra governance.

Use when OZM-governed work must keep reference projects, papers, frameworks, engines, products, or prior implementations as source-backed method anchors during later planning, dispatch, writing, review, and closeout.

Use for OZM-governed repo graph/reconstruction work: CodeGraph/codegraph MCP or CLI, .codegraph/.understand-anything, .repoanalysis bundles, graph freshness, impact radius before write, and mechanism fidelity.

Use for OZM-managed AGENTS.md, CLAUDE.md, and repository instruction surface governance.

Use before OZM-governed writer admission to analyze the request, predict landing and blockers, prepare primary and fallback implementations, freeze naming and boundaries, and build the required maps.