PI Cognitive Workflow Engine
Adds selectable reasoning modes, task-specific workflows, and verification standards to AI-assisted engineering work.
- Source repo
- share-skills/pi
- Stars
- ★ 108
- Last updated
- 4mo ago
- License
- Apache-2.0
- Primary language
- HTML
- FA score
- 34/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A developer using Claude Code can apply the programming workflow to implement changes and follow its verification checks.
- Not a fit if
- Teams using an unlisted platform that cannot adapt the installation
- Users unable to run Git and Node.js/npm for the decision visualizer
- Source review
- 34/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
PI is a set of skills and platform configurations for AI coding assistants and related workflows, with six cognitive archetypes, nine task scenarios, and lightweight, standard, and deep modes. It defines programming, testing, product, and operations workflows alongside scenario routing, task decomposition, verification guidance, and delivery quality gates. Claude Code can install it as a plugin; the README also documents skill-file setup for Codex CLI, Gemini CLI, Copilot CLI, and Qwen Code, plus rule or steering files for Cursor and Kiro. An optional local visualizer, built with TypeScript, React, and Vite, renders decision history from ~/.pi/decisions as an interactive page. PI does not document a hosted inference service; actual execution depends on the host platform, and model charges are not specified.
After installation, PI's SKILL.md or platform-specific configuration supplies the host assistant with cognitive modes, scenario routing, and task procedures. It selects a scenario and difficulty level, guides decomposition and verification, and applies delivery checks for programming, testing, product, or operations work. For multi-agent tasks, the README describes Leader, Teammate, and Coach roles and provides agents/pi-teammate.md and agents/pi-coach.md templates. The interactive install.sh detects platforms and lets users choose a language. The optional ~/.pi/visualize.sh launcher displays sanitized decision history from ~/.pi/decisions as an interactive tree, with timeline playback, archive import/export, and WebSocket live preview.
- A developer using Claude Code can apply the programming workflow to implement changes and follow its verification checks.
- A QA engineer can use the testing scenario to structure boundary coverage, execution, and failure diagnosis.
- A product manager can use the product decision scenario to break down user pain points, compare options, and define measurable indicators.
- A team coordinating multiple AI agents on cross-file work can use the Leader, Teammate, and Coach roles and templates.
- A developer reviewing how an assistant reached decisions can use the local visualizer to inspect decision trees and timelines.
How do you install or deploy this agent?
The recommended installer requires shell access and a network connection, and interactively detects platforms and offers a language choice:
curl -fsSL https://raw.githubusercontent.com/share-skills/pi/main/install.sh | bashAlternatively, install manually for a host. For Codex CLI:
mkdir -p ~/.codex/skills/pi && cp skills/pi/SKILL.md ~/.codex/skills/pi/SKILL.mdFor Claude Code, the README documents this Marketplace command:
claude install-plugin https://github.com/share-skills/piThe README does not specify model credentials or charges; you use the relevant host platform. The visualizer requires Git and Node.js/npm on first launch.
How do you use this agent?
After installation, use the host's AI assistant for a task; PI's skill or platform configuration supplies scenario, difficulty, and delivery guidance. The README does not provide a single first prompt that applies across platforms. To inspect local decision history, run:
~/.pi/visualize.shOn first launch, the visualizer can bootstrap its runtime and requires Git and Node.js/npm. If the host has the documented commands/ routing integrated, it also supports:
/pi visualizeWhat are this agent's strengths and limitations?
- Three difficulty modes combine with task scenarios and cognitive strategies to tailor the workflow to a task.
- The programming workflow specifies six debugging steps, five code-review dimensions, and build and test delivery checks.
- The README gives concrete installation paths for Claude Code, Codex CLI, Cursor, Kiro, and other named hosts, with Chinese and English variants.
- The optional visualizer supports decision trees, timeline playback, sanitized archive import/export, and WebSocket live preview.
- PI operates through host skills, plugins, or rules, so behavior and available capabilities depend on the host.
- Installation differs by platform; progressive references are explicitly supported only for Claude Code, Copilot CLI, and OpenClaw, while Cursor and Kiro use the complete single-file version.
- The visualizer needs Git and Node.js/npm on first launch and adds setup beyond installing the skill alone.
- The README does not specify model costs, required credentials, or detailed model capabilities for each host.
How does this agent compare with similar options?
The README contrasts PI with fixed-persona and single-incentive prompting and with workflows that stop after failure; it describes dynamic archetypes, escalating recovery stages, and verification-led delivery. It does not compare PI with another named product.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| PI Cognitive Workflow Engine This agent | 34 · Major gaps | Agent plugin / skill | ★ 108 | 4mo ago | HTML | Codex · Claude Code |
| Agentic Plugin Marketplace | 66 · Some gaps | Agent plugin / skillFree + model costs | ★ 42k | 3d ago | Python | Codex · Claude Code |
| EOC — Everything OpenAI Codex Workflow OS | 75 · Good | Agent plugin / skillFree + model costs | ★ 80 | 5mo ago | JavaScript | Codex |
| Spec Kit Skills for Coding Runtimes | 51 · Major gaps | Agent plugin / skill | ★ 170 | 2d ago | Shell | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
README recommends a curl | bash one-click install that writes ~/.pi/visualize.sh and bootstraps a runtime (git/node/npm) on first run — an external effect with no uninstall/rollback guidance, so rollback scores 0; least_privilege, user_confirmation, data_flow_transparency and sensitive_data_handling are only loosely asserted (e.g. the visualizer claims privacy-sanitized archives) with no concrete mechanism or confirmation flow, each 1; dependency_security has no lockfile or audit evidence, 1; source_attribution cites classical texts but not code/asset provenance, 1.
Internal inconsistency: title says v23.2.0 while the description and badge set are sprawling, and some tables contradict themselves (the 'Ten Commandments' lists only five), self_consistency 1; dependency_availability only mentions git/node/npm with no version constraints, 1; failure_messages and error handling are not evidenced in the visible files, 1.
Broad multi-platform (Claude Code, Codex, Cursor, etc.) and multi-scenario (programming/testing/product/operations) coverage supports audience_and_scenarios and environment_fit at 2; capability_boundaries are mostly marketing assertions without an explicit 'what it won't do' scope, 1; trigger_precision relies on a keyword routing table with no matching-rule evidence, 1.
Information architecture is clear (layered directories, progressive disclosure, bilingual), install_notes give per-platform steps, and license is a full Apache-2.0 text, each 2; naming_stability is at risk from version drift and pi/pi-en directory naming, 1; examples are mostly screenshots with no FAQ, 1; no known-limitations section, 0; versioning_changelog has only badges and a release workflow, no changelog, 1; maintenance_responsibility is unclear given an unverified publisher and no maintainer statement, 1.
output_usability rests on screenshots and tables without reproducible input/output samples, 1; marginal_value over generic prompt engineering cannot be confirmed statically, 1; cost_benefit claims ~43% token savings without benchmark data, 1.
Many README claims (e.g. 'tool-verified delivery', '43% token savings') have no traceable source, claim_traceability 0; the benchmark/ test files do not correspond to the PI product and cannot corroborate it, cross_source_corroboration 0; fact and inference are blended (e.g. 'real screenshots' unverifiable), fact_inference_separation 0.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- The one-click install uses curl | bash, writes into the user home, and bootstraps a runtime on demand; review install.sh and setup-standalone-visualize.sh before running and confirm how to uninstall/roll back.
- README contains self-contradictions and inflated claims (e.g. 'Ten Commandments' lists only five; token-savings ratio has no benchmark), so do not infer real effectiveness from it.
- Publisher identity is unverified and maintenance responsibility/update path are unclear; confirm maintenance status before adoption.
- The benchmark/ test files do not correspond to the PI product and cannot serve as quality evidence for it.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Full Changelog: https://github.com/share-skills/pi/compare/1.1.7...1.1.8 Full Changelog: https://github.com/share-skills/pi/compare/1.1.7...1.1.8View on GitHub ↗ -
Release notes excerpt
Full Changelog: https://github.com/share-skills/pi/compare/1.1.6...1.1.7 Full Changelog: https://github.com/share-skills/pi/compare/1.1.6...1.1.7View on GitHub ↗ -
Release notes excerpt
Full Changelog: https://github.com/share-skills/pi/compare/1.1.5...1.1.6 Full Changelog: https://github.com/share-skills/pi/compare/1.1.5...1.1.6View on GitHub ↗
FAQ
Does PI include a model or hosted inference service?
Which platforms have documented installation steps?
What does the visualizer show?
~/.pi/decisions as an interactive decision tree, with timeline playback, archive import/export, and live preview.