pi-engineer

A controlled way to delegate one clearly defined coding task to a project-scoped Pi engineer. It requires written boundaries, acceptance checks, and allowed file paths.

In plain words
What is it for?
It is for sending a new implementation block or correction to the Pi engineer, specifying required results and checks, and reviewing the returned evidence before accepting the work.
Why use it?
It separates a bounded implementation task from requirements discovery and architecture decisions, making the work easier to review and contain.

Skill for Claude CodeCodex

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 skills/catwillgh/mainframe/pi-engineer
Any agent
npx skills add CATWILLgh/MAINFRAME --skill pi-engineer
Clone the repo
git clone --depth 1 https://github.com/CATWILLgh/MAINFRAME

Made for: Claude Code, Codex.

Per session 73 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 543 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 $0.00073 $0.00543
Opus 5 $0.00036 $0.00271
Sonnet 5 $0.00015 $0.00109
Haiku 4.5 $0.00007 $0.00054

Measured 2d ago against content hash f74f3ffeb8c1, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

pi-engineer 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 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.

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.

Origin

Copies of this mod

1 near-identical copy found in the catalogue:

adapters/claude-code/plugin/skills/pi-engineer/SKILL.md · 50 lines

What it actually says

Pi engineer

Keep architecture, user communication, final review, and the commit in this primary session. Pi implements one already-agreed block in the current Git worktree and returns structured evidence; its internal verifier is a quality gate, not final acceptance.

For a new block, write a short JSON request inside .agents/runtime/pi/requests/ with this exact shape:

{
  "schemaVersion": 1,
  "goal": "One observable result",
  "writePaths": ["path/or/narrow-glob"],
  "excludePaths": [],
  "invariants": ["Behavior that must remain true"],
  "acceptance": ["Concrete result that can be checked"],
  "forbiddenFutureStages": ["Later work that must not begin"],
  "checks": [{"argv": ["exact-executable", "arg"], "timeoutMs": 60000}]
}

Keep lists only as detailed as the block requires. Checks must be exact argv, not inline shell, and cannot invoke Git. Then run mainframe-pi engineer --mode new --request <project-relative-request.json>.

For a correction to the same active block, write a correction packet under the same runtime directory and run mainframe-pi engineer --mode resume --feedback <project-relative-feedback.json>. The packet contains exactly instructions, missingEvidence, and failedCheckIds arrays. Omit --feedback only when resuming interrupted work without new review findings.

Let a run finish unless it reports a real block. Inspect the returned status, changed paths, checks, acceptance evidence, and verifier verdict against the actual diff. ready-for-architect-review means only that Pi's internal pass is complete. If review finds an in-scope defect, send one precise resume correction. If accepted, create the Conventional Commit here, limited to accepted task paths and preserving unrelated dirty or staged work. The next new block reconciles that commit and compacts the persistent Pi session.

Do not copy Pi's internal pipeline into the prompt, pass profile/config/project overrides, ask Pi to commit, or treat its verifier as user acceptance.

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 · 50 lines · 73 tokens per session scan A f74f3ffeb8c1

Subscribe to this mod's changes

pi-engineer is a skill published in the GitHub repository CATWILLgh/MAINFRAME (2 stars, last pushed 11d ago), licensed MIT. It adds 73 tokens to every session and 543 once invoked, about $0.0004 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.

Related

Other skills, from other repositories

agentforge-protocol

Use when doing non-trivial coding with Hermes, OpenClaw, Claude Code, Codex CLI, or similar autonomous coding agents. Orchestrates Karpathy-style minimal-change discipline, grill-plan intake, TDD, systematic debugging, subagent-driven implementation, spikes, and pre-commit review into one end-to-end workflow.

Yat-mo/agentforge-protocol · 71 tokens

gemini-agents-api

Manages custom Agent resources on Gemini Enterprise Agent Platform. Use when the user wants to programmatically create, configure, list, update, or delete stateful, server-managed Agent resources (including mounting files, skills, and tools) before executing conversations.

google/skills · 56 tokens

subagent-driven-development

Use when executing implementation plans with independent tasks in the current session.

obra/superpowers · 17 tokens

warp-delegate

Delegate a coding task to the Warp Agent CLI (oz) as a background implementer, then review its diff and land it yourself. Use this whenever the user wants to hand implementation work to Warp - phrasings like "have Warp implement X", "delegate this to the Warp CLI", "run it through Warp", "use oz to…

amElnagdy/delegate-skills · 144 tokens

agy-delegate

Delegate a coding task to the Google Antigravity CLI (agy) as a background implementer, then review its diff and land it yourself. Use this whenever the user wants to hand implementation work to Antigravity or agy - phrasings like "have Antigravity do X", "delegate this to agy", "run it through agy", or "use…

amElnagdy/delegate-skills · 132 tokens

zcode-delegate

Delegate a coding task to the Z.AI ZCode CLI as a background implementer, then review its diff and land it yourself. Use this whenever the user wants to hand implementation work to ZCode — phrasings like "have ZCode do X", "delegate this to ZCode", "run it through ZCode", or "use ZCode to implement/fix/refactor" — or…

amElnagdy/delegate-skills · 124 tokens