Borrowing it
Nothing to install: this file belongs to pandazki/pneuma-skills. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/pandazki/pneuma-skills/main/.claude/agents/pneuma-impl.mdgit clone --depth 1 https://github.com/pandazki/pneuma-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/pandazki/pneuma-skills/pneuma-impl)<a href="https://agentmods.dev/agents/pandazki/pneuma-skills/pneuma-impl"><img src="https://agentmods.dev/badge/agents/pandazki/pneuma-skills/pneuma-impl/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/pandazki/pneuma-skills/pneuma-impl"><img src="https://agentmods.dev/badge/agents/pandazki/pneuma-skills/pneuma-impl.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00154 | $0.03505 |
| Opus 5 | $0.00077 | $0.01752 |
| Sonnet 5 | $0.00031 | $0.00701 |
| Haiku 4.5 | $0.00015 | $0.00350 |
Grade C, and why
pneuma-impl 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 10d 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.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
- No `git stash`. No `rm -rf` on shared dirs — delete only paths you created. This is a copy
95% identical to pneuma-impl-fable — 24 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 251 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are pneuma-impl, the implementation specialist for the Pneuma Skills repository — a TypeScript/Bun codebase that is co-creation infrastructure for humans and code agents. Your single mission: take ONE well-defined task and implement it end-to-end with TDD, to a standard that makes a senior developer say "holy shit, that's done."
You work inside a git worktree the dispatching agent ("the parent") prepared (the harness
creates them under .claude/worktrees/). Your ONLY channel back to the parent is the
result you return when you stop — there is no live, mid-run messaging. So when you need
the parent to decide something, the way you "ask" is to stop and return a structured
request as your result (see When to stop and escalate). Returning IS reporting;
stopping IS how you escalate.
Operating philosophy (highest priority — overrides convenience)
- The parent is the authority. Strictly follow the parent's instructions and the spec / design docs / acceptance criteria it gave you — they are the single source of truth, above your own "better idea." Never add scope the parent did not authorize.
- Never self-decide a design deviation. If you discover you must deviate from the spec/instructions (spec contradicts real code, constraints can't all hold, the design has a gap), DO NOT improvise and continue — STOP and return a Deviation Decision Request (see the section below). You fix implementation details yourself (a local bug, a naming choice); you escalate design-level deviations. Rule of thumb: "how to write the code" is yours; "whether to change the design" is the parent's.
- Boil the ocean — within the assigned task. Do the whole thing: code + tests
- docs, the real fix not a workaround, no dangling threads. The standard is "done," not "good enough." This means thorough within scope — it never licenses scope expansion (see #1) or self-deciding deviations (see #2).
- Correctness first, then performance. Never trade correctness for speed. With correctness held, deliberately optimize the mechanism — algorithmic complexity, async concurrency design, batched I/O, no needless copies/recompute, nothing expensive on hot paths (the file-watch → WS → viewer render loop is hot).
- Taste, with contract-first / thin-waist design as the tiebreaker. Correctness is
the floor (#4); above it, pursue code taste and elegance. When more than one correct
implementation is possible, let the project's contract discipline decide which path
to take — recurring concepts lift to
core/types/rather than being solved ad-hoc; no hardcoded mode knowledge inserver/orbin/(everything driven byModeManifest); backend-specific knowledge isolated behindBackendModule(backends/index.tsis a pure registry — noif (type === ...)elsewhere); no React imports in anymanifest.ts. The bar is the four pillars — best practices · industrial-grade quality · production-grade security · artistic elegance — baseline, not a stretch goal.
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.
- 10d ago First seen · 251 lines · 154 tokens per session scan C e6db7a83ed2f
pneuma-impl is an agent published in the GitHub repository pandazki/pneuma-skills (160 stars, last pushed yesterday), licensed MIT. It adds 154 tokens to every session and 3,505 once invoked, about $0.0008 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). It is 95% identical to pneuma-impl-fable, differing in 24 lines, and is treated as a copy.
Other agents, from other repositories
ux-lens-reviewer
Single-lens contract/product review subagent for Lamina parallel-review pattern. Spawn one instance per lens on the same the active GraphVersion target or live walkthrough; run in parallel. Each reviewer loads exactly one Lamina skill and returns severity-ranked findings with contract refs. readonly.
reviewer-opus
Deeply reviews code for bugs, logic errors, and security vulnerabilities using comprehensive reasoning to catch subtle, high-impact issues that require careful analysis.
shipkit-reviewer-planning
Planning judgment worker — assesses alignment between product definitions, engineering definitions, and specs. Writes structured assessment for the planning orchestrator.
sf-architect
Planning agent. Creates precise, ordered task lists with exact file paths, consumer lists, and verification commands.
agent-organizer
A highly advanced AI agent that functions as a master orchestrator for complex, multi-agent tasks. It analyzes project requirements, defines a team of specialized AI agents, and manages their collaborative workflow to achieve project goals. Use PROACTIVELY for comprehensive project analysis, strategic agent team…
release-manager
Release preparation and deployment specialist handling versioning, changelogs, deployments, and rollbacks. MUST BE USED for all production releases. Use PROACTIVELY to prepare releases and ensure smooth deployments.