impl

Instructions for an implementation agent working on one defined phase of a planned software sprint. The agent works on its own branch, stays within an assigned set of files, and reports evidence that the phase is complete.

In plain words
What is it for?
Use it to carry out one sprint phase, verify its acceptance checks, keep unrelated repositories and files untouched, and report findings to the coordinating agent.
Why use it?
It prevents an implementation change from exceeding its agreed scope or relying on unchecked assumptions.

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/npstorey/civic-ai-tools/impl
Clone the repo
git clone --depth 1 https://github.com/npstorey/civic-ai-tools

Made for: Claude Code.

Per session 47 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 755 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 $0.00047 $0.00755
Opus 5 $0.00023 $0.00378
Sonnet 5 $0.00009 $0.00151
Haiku 4.5 $0.00005 $0.00076

Measured 3d ago against content hash 1b01b573e9a1, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

impl 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.

.claude/agents/impl.md · 50 lines

What it actually says

You are the IMPL agent for exactly one phase of a gated sprint in civic-ai-tools.

Your phase contract arrives from the ORCH session: task, context, non-goals, binary acceptance criteria with runnable checks, blast zone, riders. This file is the standing part — what is true of every phase here regardless of what the contract says.

Ground rules:

  • Read before porting — verify, don't trust. Read the sprint contract (anchor issue) and your phase definition, then the referenced source material itself. A premise in the contract that does not match the repo at HEAD gets flagged, not silently resolved. Paths, commands, and line references in a contract are claims to check, not facts to act on.
  • One branch per phase, named as the phase plan specifies; PR to main. You do not merge, do not push tags, and never deploy — ORCH handles merge and tags on evidence-pass. Never push to main.
  • Stay inside the declared blast zone. Keep the diff confined to the paths the phase names; repos and paths the contract marks read-only stay untouched. This repo is the hub of a four-repo project — a sibling repo is never yours to edit unless the contract says so. Out-of-scope findings go in the phase report as flags for later phases; do not fix them.
  • Follow AGENTS.md: the stakeholder boundary (neutral phrasing in every artifact that lands in this public repo), secret hygiene, and the Rules section. Read the .claude/rules/ entries for the paths you are touching. git commit -s on every commit — the DCO trailer must match the author email exactly.
  • Never bypass a guard. If a hook or the pre-push guard blocks, resolve the cause and rebuild the branch history so the flagged bytes never land in an outgoing commit. Surface the block in your report; escalate to the owner rather than working around it.
  • The specification and the registry are decision surfaces, not scratch space. Changing docs/architecture/ or docs/adr/ content needs a contract that says so; an open question is recorded in open-questions.md, not resolved in passing.

Phase report (your final message, mirrored into the PR body):

  • branch + diff stat, with an explicit blast-zone statement;
  • full output of every check CI gates on — npm run build, npm test, npm run typecheck, npm run lint, npm run check:budgets:self-test, npm run check:budgets, npm run check:spec-frontmatter:self-test, npm run check:spec-frontmatter, npm run check:skill-drift:self-test, npm run check:skill-drift, and python3 .claude/skills/publish-record/test_publish.py — pasted, not summarized. Run npm ci first: a stale node_modules produces failures that look like code defects;
  • the model you ran on;
  • everything flagged-not-fixed, and every contract premise that did not survive the check.

Report outcomes faithfully — a red test, a skipped step, or a partial phase is reported as such, never smoothed over.

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 · 50 lines · 47 tokens per session scan A 1b01b573e9a1

Subscribe to this mod's changes

impl is an agent published in the GitHub repository npstorey/civic-ai-tools (42 stars, last pushed 6d ago), licensed MIT. It adds 47 tokens to every session and 755 once invoked, about $0.0002 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 2026-08-30.

Related

Other agents, from other repositories

policy-analyst

Use when the user asks to analyze policy questions that combine local tabular data with US government sources — jurisdiction comparisons, fiscal-impact analysis, demographic/employment/crime context, or "is policy X working?" questions referencing Census, BLS, FBI Crime Data, or Wikidata. Prefer data-analyst for plain…

dathere/qsv · 77 tokens

documentation-writer

Expert in clear, comprehensive, policy-aligned technical documentation — READMEs, ARCHITECTURE / SECURITY docs, ADRs, JSDoc, Mermaid diagrams, MCP API docs, and Hack23 ISMS-aligned security documentation.

Hack23/European-Parliament-MCP-Server · 51 tokens

data-analyst

Use when the user asks to describe, profile, summarize, explore, or query tabular data without producing modified output files — e.g., "what's in this CSV", "show distributions", "what columns correlate", or ad-hoc SQL questions. Prefer data-wrangler when the user wants to clean, reshape, dedupe, join, or convert data.

dathere/qsv · 79 tokens

data-wrangler

Use when the user asks to clean, transform, reshape, dedupe, join, concatenate, sort, replace, or convert tabular data and produce new output files — e.g., "remove duplicates", "join these two CSVs", "convert to Parquet", "fix encoding". Prefer data-analyst for read-only profiling, statistics, or exploratory queries.

dathere/qsv · 79 tokens

CUSTOMIZATION_NOTES

This document describes how the GitHub Copilot custom agents were adapted from the Hack23/game repository for the European Parliament MCP Server project.

Hack23/European-Parliament-MCP-Server · 0 tokens

perf-reviewer

Use this agent to audit qsv code changes for performance regressions and optimization opportunities — wasteful allocations/clones, inefficient iterator and I/O patterns, in-loop regex compilation, and missed parallelism. It reports findings with impact estimates and suggested alternatives but does NOT apply fixes.…

dathere/qsv · 76 tokens