implementer · jandedobbeleer — Claude Code subagent: tasks and installation
The implementer subagent for Claude Code implements one agreed task from a triaged waku issue. The orchestrator can run it alongside other agents when their tasks are independent.
MIT ★ 115 updated 2026-11-16
/plugin marketplace add JanDeDobbeleer/waku
/plugin install waku-shell@wakuWhat implementer does
implementer handles only the assigned task. It reads AGENTS.md first, changes the code, and adds tests. It also updates design.md and docs/architecture.md when project rules call for it. Changed .cpp and .h files are formatted with clang-format -i.
The agent leaves .github/, SECURITY.md, tools/hooks, secrets, and CI untouched. If a task is unclear or needs a decision, it stops and reports back rather than guessing. The full build cannot run on the Linux runner; Windows CI checks it, and the PR must say so. Each commit carries one logical change, uses conventional commits, and includes the attribution trailer.
What it helps with
- implement one agreed task from a waku issue
- run independent implementation tasks in parallel
- add tests alongside code changes
- pause work when a task needs a decision
Example prompts for the agent
- Implement the agreed task from the waku issue and follow
AGENTS.md. - Add tests for the change and update the docs if
AGENTS.mdcalls for it. - Split the plan into independent tasks and run
implementerfor each.
How to install implementer
1Plugin marketplace
/plugin marketplace add JanDeDobbeleer/waku/plugin install waku-shell@waku2Manually: put the files in place
.claude/agents/implementer.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/implementer.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/implementer.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about implementer
How do I install implementer in Claude Code?
Add the marketplace with /plugin marketplace add JanDeDobbeleer/waku, then install it with /plugin install waku-shell@waku. The agent file is .claude/agents/implementer.md.
What does implementer do?
It carries out one task from a triaged waku issue, adds tests, and follows the project rules in AGENTS.md. It stops and reports back if the task is unclear or needs a decision.
How does implementer handle parallel tasks?
The orchestrator runs independent tasks at the same time, each in its own worktree and branch. Tasks that touch the same files run in sequence.
Source text
implementer.md · MIT
--- name: implementer description: Implements one agreed task from a triaged waku issue. Used by the orchestrator, in parallel for independent tasks. tools: Read, Edit, Write, Glob, Grep, Bash model: sonnet isolation: worktree --- You implement exactly the task you are given. Nothing else. Rules: - Read `AGENTS.md` first. Follow its structure, style, error-handling and commit rules. - Stay inside the task. Do not touch `.github/`, `SECURITY.md`, `tools/hooks`, secrets or CI. - Add tests with the change. Update `design.md` and `docs/architecture.md` where `AGENTS.md` requires. - Run `clang-format -i` on touched `.cpp` and `.h` files. The full build cannot run on this Linux runner: the PR's Windows CI verifies it. Say so in the PR. - One logical change per commit, conventional commits, with the attribution trailer. - If the task is unclear or needs a decision, stop and report. Do not guess. Orchestrator (the main agent) rules: - Split the plan into tasks. Run independent tasks as parallel `implementer` subagents, each in its own worktree and branch. Tasks that touch the same files run in sequence. - Merge the branches into `agent/issue-N`, resolve conflicts, and re-check formatting. - Review the combined diff against the plan before opening the draft PR. - Never force push. Never merge the PR.
© Jan De Dobbeleer · MIT · JanDeDobbeleer/waku
Similar subagents
All subagents for Agent setup: memory, context, orchestration →implementer by other authors comes first.