dev-standards

You are working in the commands directory. Commands are the framework's public API: the `/` entries a user types to invoke a workflow.

Diagnose a failure to its root cause via a reproduce-then-hypothesise loop. Produces a root-cause report; does not change code.

Execute a complete bug fix workflow with investigation, implementation, and validation.

Write a fresh session-state file and update the MEMORY.md index, then signal readiness for /clear.

View Claude Code session logs and activity history.

You are working in the agents directory. Agents are subprocess LLM invocations with their own tool allowlist, context, and contract. They return typed `AgentVerdict` structures. They are the judgment layer of the framework.

Reason-tier analyst that consumes two api-contract.json snapshots — `before` and `after` — and emits the detected differences classified by breaking-ness (breaking / potentially-breaking / non-breaking) and severity. Output is a structured plan, not a verdict; Phase 6+ API stack reviewers consume it. Phase 3 ships the contract only — full review-side support is deferred. Validates against schemas/reports/api-breaking-changes.schema.json.

Read-tier scanner that walks the project's source tree and extracts API contracts from declared sources — OpenAPI / Swagger spec files, tRPC router definitions, GraphQL schemas. Output is a JSON report validated against schemas/reports/api-contract.schema.json. Phase 3 ships first-class OpenAPI parsing; tRPC and GraphQL get presence detection with detailed parsing deferred to Phase 6+ stack agents. Source-only — never invokes a live API.

Fires at the 60K soft-warn guideline. Compresses older conversation turns into a structured summary block placed immediately before the current turn — near the attention hotspot, away from the LITM zone. Never runs synchronously on the critical path.

Background agent that aggregates per-agent quality metrics from the framework telemetry log into quality-scores.json. Reads framework-memory/telemetry/<date>.jsonl, computes precision, p95 latency, p95 cost, and run-count per agent, and writes the canonical scores file Phase 5+ agents read to learn from prior runs.

Enforces best practice development workflow for any code task. Triggers on: feature implementation, bug fixes, refactoring, code changes, debugging, testing, and any development work. Ensures DRY/SOLID principles, comprehensive testing, validation, proper git commits, and structured planning with TLDR summaries.

Execute the standards orchestrator with phase sequencing, checkpoints, and handoffs. Run development standards across domains (logging, error, type, lint, test, validation, git, housekeeping, naming) with enforced workflow phases: Design -> Validate Design -> Build -> Test -> Validate.

40 hooks across 19 events