terylon-dev

Plugin terylon-dev from marketplace terylon in janecekvit/terylon-marketplace. Commands read from the repo's manifest — today.

Skills (5)

• background-run

  Use when the operator asks for work to leave this session and run on its own — "run this in the background", "dispatch this as a background agent", "kick this off overnight", "start it while I'm away", "run it unattended". Dispatches it with claude --bg into a worktree whose name is chosen rather th

• brainstorm

  Use when an idea is still too vague to plan — "I want to build X", "we should probably…", a half-formed feature, or a work item whose description only restates its title. Grounds the idea in the codebase, runs a one-question-at-a-time dialogue, writes a design document, and decomposes it into seed-s

• develop

  Use when implementing one or more user stories end-to-end — give work-item or issue URLs from either forge, a prose description, or several prose items separated by --- on its own line. Several items fan out into concurrent leaders, one workspace each. Hosts the human gates and dispatches the leader

• run-build-loop

  Use when executing an approved implementation plan task by task with a fresh implementer subagent per task. Defines the ledger that survives compaction, the pre-flight scan of the plan, the bounded fix-round loop after a review, and how to pick a model per role. For the controller of the loop, not f

• write-plan

  Use when turning an approved spec or seed-spec into an implementation plan that a per-task implementer will execute without the context that produced it. Defines the plan document's structure — header, global constraints, file map, task shape, interfaces — and the self-review that runs before the pl

Subagents (9)

• architecture-reviewer

  Use after a task's tests are green to review the diff through an architecture lens: module boundaries, dependency direction, consistency with surrounding patterns, placement of responsibilities. Read-only — reports findings, never edits.

• debugger

  Use when the same test is still failing after the implementer has already attempted it twice — not only when it reports BLOCKED: find the root cause systematically (hypothesis, reproduction, minimal fix) instead of letting the implementer keep guessing in a context that is already large. Read-only o

• developer

  Use as the per-task implementer in a build: implement ONE plan task test-first (TDD red-green-refactor) following the target repo's conventions, run the covering tests, commit, and self-review. Bounded — hands over at a turn budget rather than running a whole task in one context. Reports DONE / DONE

• edge-case-reviewer

  Use after a task's tests are green to harden coverage: find untested edge/negative/error paths and propose the missing test cases for the developer to add (still test-first). Read-only — proposes tests, never edits code.

• leader

  Use as the autonomous build controller: receives an approved seed-spec or plan path, dispatches planner / developer / debugger / refactorer / review lenses, reads their reports, and decides the next round. Returns CONTINUE / AWAITING_APPROVAL / NEEDS_CLARIFICATION / BUILD_COMPLETE / BLOCKED. Ends at

• performance-reviewer

  Use after a task's tests are green to review the diff through a performance lens: algorithmic complexity, N+1 queries, redundant allocations, blocking I/O on hot paths. Read-only — reports findings, never edits.

• planner

  Use to turn an approved seed-spec into a design + bite-sized TDD plan for a Terylon dev task. Grounds in the codebase via the Explore agent, drafts the architecture via the built-in Plan agent, and emits the artifacts in the write-plan format. Returns READY_FOR_BUILD or NEEDS_CLARIFICATION; never gu

• refactorer

  Use after a task's tests are green to simplify the diff: remove duplication, reuse what already exists, align with surrounding conventions. May edit, but only while the tests stay green. Reports SIMPLIFIED or NO_CHANGE.

• security-reviewer

  Use after a task's tests are green to review the diff through a security lens: injection, authn/authz, data exposure, input validation, secret handling. Read-only — reports findings, never edits.

Similar plugins