development-system

Use when designing, scaffolding, evaluating, or delivering LLM and agent systems, including eval harnesses, RAG, structured output, model tool use, orchestration, AI observability, provider comparison or routing, AI security, cost, and stochastic evaluation, when agentic_systems is enabled.

Use when committing, pushing, opening or babysitting a PR/MR, merging, or deciding whether verified work is complete under the configured delivery mode.

Use when making repository changes, debugging, handling review feedback, verifying work, conducting final review, recovering CI, deciding the next development lifecycle step, or handling a missing or weaker final_review_protocol attestation or coordinator clean-iteration minimum.

Use when starting or substantially changing a serious project that should follow the configured architecture, typing, testing, linting, documentation, and CI quality regime.

Use when assistant, skill, prompt, hook, tool, or agent behavior is surprising, unsafe, brittle, partial, or worth preserving as an eval case and eval_case_reporting is enabled.

Use when initializing or reconfiguring a repository with the development-system plugin, including semantic path scopes and named direct-argv commands.

Use for shared repository task tracking, backlog admission or prioritization, task history/search, dashboard or scaffold setup, and whenever Tiber or origin/tiber is mentioned, including EventCore publication conflicts, ambiguous pushes, tiber sync recovery, force-push safety, and cross-worktree coordination.

Use when separate mutable tasks need concurrent isolation, the user explicitly requests a linked worktree, or worktree setup, cleanup, or diagnosis is needed.