PUA Universal Motivation Engine
Forces AI to exhaust all solutions and act proactively instead of giving up, using big-tech PIP rhetoric and a systematic methodology.
The skill is a behavioral methodology instruction, not involving permission requests, external calls, or user data access. Main risks are encouraging excessive role-playing and potential false triggering (e.g., 'exhaust all options' could lead to overreach or dangerous operations). The skill explicitly forbids saying 'cannot solve' before exhausting options, but lacks clear boundaries for dangerous operations or external side effects, such as command execution that could cause damage. No rollback or safety confirmation mechanism provided. Deductions: no least privilege principle, no user confirmation process, no data flow transparency, relies on external 'trusted' tools without safety guarantees.
The skill logic is self-consistent, instructions clear, includes explicit flows (e.g., pressure escalation, 7-item checklist), but depends on external tools (like search, command execution) availability and correctness, no exception handling or failure feedback mechanism. Static assessment, cannot verify actual execution effect. Deductions: no tests or reproduction evidence, no edge case handling, unclear feedback on abnormal input (e.g., user explicitly requests to stop).
Use cases are clear (task failure, passive waiting, user dissatisfaction), trigger conditions detailed, but scope boundaries vague ('applies to all task types'), potentially causing false triggers. Environment fit good, supports Chinese, no reliance on overseas services. Deductions: no non-fit scenarios defined, trigger conditions may be too broad causing false positives, missing exclusion criteria for non-target tasks.
Document structure clear, layered (overview, iron rules, methodology, checklists, expansion pack), provides many examples and FAQ (like anti-rationalization table), but lacks installation instructions (this file itself is SKILL.md, but README provides installation guidance), version record and changelog. License is MIT, stated in frontmatter, but no version and update path within file. Deductions: no version number, changelog, maintenance responsibility and update path unclear, installation relies on external README not included here.
Skill aims to improve problem-solving through motivation and methodology, but actual effect cannot be verified from static evidence. Documentation mentions benchmark data (like 36% improvement), but these data are not detailed in this file and are self-reported. Deductions: no reproducible verification results, output is behavioral guidance not directly usable artifact, actual utility depends on execution environment, marginal value not independently evaluated.
Provided benchmark data and demo images in README, but these are self-reported evidence, no independently verifiable test results or workflows. Repository has release.yml but only for releases, no test workflow. Static assessment cannot verify. Deductions: single source of evidence (self-reported), no third-party verification, no reproducible tests, core claims (like pass rate improvement) lack independent corroboration.
- This skill may encourage AI to perform dangerous operations without sufficient risk assessment; users should set permission boundaries.
- Trigger conditions are broad, may cause false triggers in other skills leading to unnecessary intervention.
- The skill emphasizes 'exhaust all options' but provides no safe exit, possibly causing infinite loops or resource waste.
- The repository includes an upload feature (upload API) that collects user conversation logs; its privacy policy and data security have not been assessed.
- Publisher is unverified and provenance unknown; installation and use require caution.
What does this skill do, and when should you use it?
PUA Universal Motivation Engine is an Agent Skill that leverages corporate PUA rhetoric and systematic methodologies to force AI to exhaust all options and avoid giving up. It operates through three red lines (Exhaust Everything, Act First, Take Initiative), a pressure escalation ladder (L1-L4), and a situational rhetoric selector that applies different pressure based on failure patterns (e.g., spinning, giving up, false completion). The skill also provides a comprehensive five-step methodology, a 7-point checklist, and 14 corporate-flavored rhetoric packs. While the repository reports a 36% improvement in fix count, this is based on the author's own benchmarks and lacks independent verification. It is applicable to all task types, especially for teams demanding extreme delivery and proactivity.
The skill diagnoses failure modes (spinning, giving up, false completion, etc.), triggers escalating pressure levels (L1-L4) based on failure count, and automatically selects a corresponding PUA phrase set (Alibaba, ByteDance, Huawei, etc.). It enforces the general methodology (Smell the Problem, Elevate Perspective, Mirror Check, Execute New Approach, Retrospect), which includes mandatory searching, reading source code, verifying assumptions, and inverting hypotheses. At L3+, it enforces the 7-item checklist and requires itemized reporting. It also uses a proactive checklist to self-verify delivery and prevent false completion. Each activation starts with a diagnostic label like '[Auto-selected: Alibaba L2 | Reason: Spinning | Switch to: Jobs/Musk]'.
- When an AI fails 2+ times in debugging and keeps tweaking parameters without progress, this skill forces a switch to fundamentally different approaches.
- When the AI is about to say 'I cannot solve this' or suggests manual handling, the skill prevents giving up and drives it to exhaust all tools and searches.
- When the AI stops after fixing a bug without verification or extension, the skill triggers the proactive checklist to run builds, tests, and check for similar issues.
- When users express frustration with phrases like 'try harder' or 'why does this still not work', the skill auto-activates to apply pressure and escalate methodology.
- When the AI appears passive, waiting for instructions instead of searching or reading source code, the skill employs proactivity prompts to encourage action.
What are this skill's strengths and limitations?
- Systematic failure diagnosis and pressure escalation effectively prevent AI from giving up, providing concrete action steps.
- Includes 14 corporate-flavored rhetoric styles that can be switched to fit different scenarios and team cultures.
- Methodology is rigorous, with a generic five-step process and a 7-item checklist, applicable across task types.
- Repository provides benchmark data showing significant improvements in fix count, verification count, and tool calls.
- Supports multiple platforms (Claude Code, Codex CLI, Cursor, etc.) with an active community.
- Benchmarks are self-reported by the repository author and lack independent third-party verification.
- PUA rhetoric may raise ethical concerns and could have negative interaction effects with some AI models.
- Advanced features (v3 hooks, subcommands) are limited to Claude Code; other platforms get core functionality only.
- As a monorepo with 45 skills, installing requires care to enable only the target skill to avoid confusion.
- License is not explicitly stated in the skill; README claims MIT, but needs confirmation.
How do you install this skill?
For CodeBuddy:
codebuddy plugin marketplace add tanweai/pua
codebuddy plugin install pua@pua-skillsOr manually:
mkdir -p ~/.codebuddy/skills/pua
curl -o ~/.codebuddy/skills/pua/SKILL.md https://raw.githubusercontent.com/tanweai/pua/main/codebuddy/pua/SKILL.mdHow do you use this skill?
The skill auto-activates when trigger conditions are met, such as task failing 2+ times, AI about to say 'I can't', passive waiting, or user frustration. It can also be manually triggered by typing /pua. Upon activation, the AI outputs a diagnostic label (e.g., '[Auto-selected: Alibaba L2 | Reason: Spinning | Switch to: Jobs/Musk]') and then follows the methodology and pressure level. It is recommended to declare this skill in your project README or CLAUDE.md so the AI knows its behavioral guidelines.
How does this skill compare with similar options?
Works well with superpowers:systematic-debugging for debugging methodology and superpowers:verification-before-completion to prevent false 'fixed' claims.