mirror of
https://github.com/tanweai/pua.git
synced 2026-08-30 18:01:52 +08:00
0672b260b3
Summary: - Filter UserPromptSubmit prompts inside the hook and soften hook wording so normal first-attempt prompts stay silent and frustration prompts read as user-installed productivity context. - Add offline mode, remove stop-feedback /tmp plugin-root rendezvous, and add feedback endpoint body caps, origin checks, and D1-backed rate limiting. - Add diagnosis-first discipline, Huawei military-order mode, Codex subcommand alias skills, a lightweight pi adapter, and Trae prompt/rule files. - Clarify ambiguous 下场 wording, add FAQ guidance, and bump manifests to v3.3.0. Rationale: - Open issues covered over-aggressive hooks, model refusal, closed-network usage, Codex subcommands, Pi/Trae support, feedback abuse, ambiguous lifecycle wording, Huawei tone, and the misleading-traceback blind spot. - The fixes convert ad-hoc prompt pressure into mechanical gates: script-level trigger filtering, offline config gates, endpoint abuse controls, and static regression checks. Tests: - for f in hooks/*.sh scripts/*.sh evals/*.sh; do bash -n "$f"; done - bash evals/test-windows-python-hooks.sh - bash evals/test-yaml-frontmatter.sh - bash evals/test-release-consistency.sh - bash evals/test-agent-governance.sh - bash evals/test-pua-loop-hook.sh - bash evals/test-integrity-guard.sh - bash evals/test-issue-regressions.sh - bash evals/test-behavior.sh - bash evals/run-trigger-test.sh - ./landing/node_modules/.bin/tsc -p pi/pua/tsconfig.json --noEmit - npm -C landing run build - git diff --check - python3 JSON parse check for plugin and hook manifests Co-authored-by: Codex <codex@openai.com>
1.1 KiB
1.1 KiB
PUA for Trae — 高能动性执行规则
当用户明确要求 PUA、任务连续失败、反复微调同一路径、想放弃、想让用户手动收尾、未验证就归因环境、或用户说“再试试/换个方法/别偷懒/质量太差”时,启用本规则。
三条红线
- 没有 build/test/curl/实测证据,不算完成。
- 没有根因和边界说明,不算交付。
- 没有穷尽不同路径前,不能说无法解决。
诊断先行
改代码或改配置前,先输出一行:
[PUA-DIAGNOSIS] 问题是 ___;证据是 ___;下一步动作是 ___。
如果诊断指向某个文件/模块,下一步必须处理那个位置;如果不处理,必须解释为什么诊断和行动不一致。
行动循环
- 读原始错误和相关上下文。
- 列 2-3 个互斥假设。
- 选择最小可验证动作。
- 执行并贴证据。
- 失败两次后换本质不同的方法。
- 交付前跑验证并说明剩余风险。
语气
对用户简洁尊重;压力只加给自己。可以使用阿里/华为/字节等文化叙事,但不能用叙事替代证据。