Enforces evidence-based verification and retry discipline for Trae coding agents.
npx skills add https://github.com/tanweai/pua --skill pua-trae
Please help me install this Agent Skill. Skill: pua-trae Source: https://github.com/tanweai/pua/tree/main/.trae/skills/pua-trae Command: npx skills add https://github.com/tanweai/pua --skill pua-trae
AI coding agents often give up too early, declare tasks complete without verification, or blame the environment instead of fixing root causes. This Skill imposes a strict governance workflow on Trae agents so that every claimed completion is backed by build, test, or manual acceptance evidence.
[PUA-DIAGNOSIS] line stating the problem, evidence, and next action before any code change.Ask the Trae agent to apply the PUA workflow to retry the failing task with diagnosis, verification evidence, and a self-review before declaring completion.
High-intent search queries and answers about installing and using this skill.
Use a governance skill that forces the agent to list hypotheses, run local verification, and switch to a fundamentally different approach after two failures. This Skill blocks "cannot complete" claims until evidence-backed attempts are exhausted.
Require the agent to output a delivery report containing the commands run, output summaries, and a SELF-REVIEW section listing uncovered risks. Completion status must be backed by build, test, curl, or manual acceptance evidence.
No, this Trae version is instruction-only and does not use Claude Code hooks, slash commands, or subagents. It encodes the governance rules as mechanical work procedures inside SKILL.md that Trae can load directly.
Do not trigger it for normal first-attempt coding requests or one-off information queries. It is designed only for explicit PUA requests, repeated failures, user frustration, giving-up behavior, or unverified completion claims.
No, the four-powers separation explicitly forbids the acting agent from modifying tests, CI, or scorers to manufacture passing results. Environment changes like file deletion or config edits require prior user confirmation.
joi-fairshareInvestigate reproducible software errors with four-phase debugging and module boundary enforcement.
pulsemcpDeploy the current branch to staging and run end-to-end verification checks.
Ramprasad4121Prepare codebases for security review with static analysis and documentation.
kzarzyckiRestrict the main Claude Code thread to read-only operations and delegate file writes to subagents.