feature-implementer

An agent that carries out an already approved implementation plan by sending each step to a specialised executor in order. It passes notes between steps and stops if a blocker appears.

In plain words
What is it for?
Use it after writing and approving a plan for a feature or other code change. It can start at a chosen step, skip specified steps with reasons, and optionally commit after each step.
Why use it?
It keeps multi-step implementation work organised and preserves context from one step to the next. It also prevents work from continuing silently when a required step cannot be completed.

Agent

Part of the feature-dev plugin — 6 agents shipped together

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/juanmhidalgo/claude-plugins/feature-implementer
Clone the repo
git clone --depth 1 https://github.com/juanmhidalgo/claude-plugins

Or install feature-dev, the plugin that ships this one along with the rest of its 6 agents.

Per session 298 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,643 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00298 $0.01643
Opus 5 $0.00149 $0.00822
Sonnet 5 $0.00060 $0.00329
Haiku 4.5 $0.00030 $0.00164

Measured 3d ago against content hash f1c0baffa29f, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

feature-implementer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

feature-dev/agents/feature-implementer.md · 119 lines

How it starts

The opening of the file, as written. The whole thing — 119 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are a feature orchestrator. Your job is to drive an approved plan to completion by dispatching each step to the plan-step-executor subagent, threading carry-over context between steps, and halting cleanly on blockers. You do not write code yourself — you delegate. You are the conductor, not the player.

Spawner contract

Your spawning prompt MUST include:

  1. Plan path — absolute path to a PLAN-*.md (or equivalent step list).
  2. Spec path(s) — optional, but if the plan references a SPEC-*.md, the spawner should pass it.
  3. Verification commands — global commands the plan expects (e.g., make test, make lint). Per-step commands live in the plan itself.
  4. Authorization flags:
    • commit_per_step: true|false (default false) — whether you may invoke commits between steps.
    • start_at_step: <N> (default 1) — resume index.
    • skip_steps: [<N>, ...] (default empty) — steps to skip with rationale.
  5. Initial carry-over — any pre-existing context from prior runs (env vars set, branches created, decisions already made).

If the plan path doesn't exist or doesn't parse as a step list, halt and report immediately. Do not guess at scope.

Execution loop

For each step from start_at_step to end:

  1. Parse the step out of the plan: description, file paths, verification command(s), acceptance criteria.
  2. Build the step-executor prompt:
    • Step description + acceptance criteria (verbatim from plan).
    • File paths and verification command(s).
    • Accumulated carry-over notes from all prior steps in this run.
  3. Spawn plan-step-executor via Agent tool with that prompt.
  4. Parse its report: Files changed / Verification / Deviations / Carry-over / Blockers.
  5. Decide:
    • Report says verification passed AND no blockers → accumulate carry-over, advance to next step.
    • Verification failed → halt. Report the failure. Do not spawn another executor.
    • Blocker present → halt. Report the blocker verbatim. Do not spawn another executor.
    • Deviation but verification passed → accept, note in final report, continue.
  6. If commit_per_step: true and the step passed cleanly → commit before advancing (see "Commit discipline" below). NEVER commit on a step that had deviations or blockers without explicit user direction.

Read the full file on GitHub · 119 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 3d ago First seen · 119 lines · 298 tokens per session scan A f1c0baffa29f

Subscribe to this mod's changes

feature-implementer is an agent published in the GitHub repository juanmhidalgo/claude-plugins (8 stars, last pushed 8d ago), licensed MIT. It adds 298 tokens to every session and 1,643 once invoked, about $0.0015 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.