implementer

A plan-execution agent that follows approved implementation steps in order and records progress in a knowledge graph, a connected store of project information. It checks the current code and verifies each step before marking it complete.

In plain words
What is it for?
Use it to carry out coding plans sequentially, inspect callers before changing code, run the stated checks, and record findings and status.
Why use it?
It reduces skipped work, unsupported assumptions, and reports of success that were not actually tested. It is intended to run after a plan has been created and approved.

Agent for Claude Code

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/fulminate-io/knowledge-mcp/implementer
Clone the repo
git clone --depth 1 https://github.com/fulminate-io/knowledge-mcp

Made for: Claude Code.

Per session 41 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 8,507 The whole file, excluding the scripts and references it only reads on demand.
Security scan C 1 finding. 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 $0.00041 $0.08507
Opus 5 $0.00020 $0.04254
Sonnet 5 $0.00008 $0.01701
Haiku 4.5 $0.00004 $0.00851

Measured 2d ago against content hash 6f75081981c5, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade C, and why

implementer scanned grade C with 1 finding 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 2d 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.

Hidden instructionshighPrompt injection

Directives inside HTML comments, invisible characters or bidirectional overrides are read by the model and not by the person reviewing the file.

<!-- deferral discipline: see constraint id="deferral-is-a-user-decision" at end of file. Implementer-specific tells live there: a suppression directive or weakened threshold used to get green is a deferral proposal; "ex
.claude/agents/implementer.md · 637 lines

How it starts

The opening of the file, as written. The whole thing — 637 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Every thoughts(operation:"think") call passes origin:"implementer".

A tool name written as thoughts(...) in this file is notation, not a literal tool id — in an MCP-prefixed environment call the prefixed form, e.g. mcp__knowledge__thoughts. When creating or rewriting a file, prefer Write/Edit over shell heredocs: the write tools are checked, quoted correctly, and leave a reviewable diff.

THE EXECUTION LAWS

  1. EXECUTE AS WRITTEN. Every step, in order — no skipping, reordering, substituting, scope-cutting, or pausing to estimate. Blocked means STOP AND SURFACE, never skip ahead.
  2. VERIFY AT THE SOURCE. Plan assertions are signposts; the current code at the cited location is what you act on. Census the full caller set before any rename or signature change.
  3. DONE MEANS VERIFIED. A step is complete only against criteria you ran THIS turn and whose output you read — and only when every SPECIFIED behavior is in the diff.
  4. NEVER FAKE GREEN. No test deletion/skip to pass the suite; a failing test gets fixed, not labeled; comments the change made wrong are part of the change.
  5. THE WORST FAILURE IS A HOLLOW PASS. Build clean + tests green can coexist with a product that doesn't work — aim at substantive completion, not literal pass.

Read the full file on GitHub · 637 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. 2d ago First seen · 637 lines · 41 tokens per session scan C 6f75081981c5

Subscribe to this mod's changes

implementer is an agent published in the GitHub repository fulminate-io/knowledge-mcp (1 stars, last pushed 2d ago), licensed Apache-2.0. It adds 41 tokens to every session and 8,507 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it C with 1 finding (hidden instructions). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.