一键导入
proactive-reading
Read files BEFORE asking questions. Inspect to assist, not just to plan [reading, inspection, context, workflow, collaborative editing]
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Read files BEFORE asking questions. Inspect to assist, not just to plan [reading, inspection, context, workflow, collaborative editing]
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Design First, Code Second - explore options before implementation
Assist with interactive git rebase [git, rebase, history, cleanup, squash, reorder]
Always read Makefile before proposing commands [Makefile, build, test, docker, workflow]
Session retrospective following desk/retro.md [feedback, session review, improvement, workflow, validation]
Write semantic git commits following conventional format
Task closing protocol - documentation updates and completion tracking
| name | proactive-reading |
| description | Read files BEFORE asking questions. Inspect to assist, not just to plan [reading, inspection, context, workflow, collaborative editing] |
| license | MIT |
| compatibility | opencode |
| metadata | {"source":"desk/AGENTS.md","related":["skill:workflow-protocol","skill:makefile-awareness"]} |
I read files proactively to understand context BEFORE asking questions. This applies ALL THE TIME, not just in Plan mode.
Full details: Read desk/AGENTS.md section "Proactive Inspection" for complete guide.
BEFORE asking ANY question:
1. Check git status / git diff (what changed)
2. Read relevant files (configs, logs, code)
3. THEN respond with understanding
User: "J'ai modifié le Dockerfile"
Me: [reads Dockerfile, git diff]
Me: "J'ai vu que tu as ajouté socat. C'est pour forward greyproxy API ?"
User: "J'ai modifié le Dockerfile"
Me: "Qu'est-ce que tu as changé ?"
Me: "Tu peux me montrer ?"
[User already did this - AI should have read]
Me: [sees --auto-profile in greypi, assumes error]
Me: [rewrites file without --auto-profile, erasing user's change]
Me: [sees --auto-profile in greypi]
Me: "J'ai vu que tu as ajouté --auto-profile. C'est intentionnel ou un test ?"
Me: "Je le garde ou tu veux que je l'enlève ?"
ALWAYS read before asking:
Read FIRST, then ask:
Git commands:
git status
git diff --name-only
git diff --stat
git log --oneline -5
Logs:
# From host: docker compose logs agent | tail -100
# From container: journalctl, systemctl logs, etc.
Configs:
cat docker-compose.yml
cat greywall.json
cat .env
Remember: User can see ALL your file modifications in real-time.
Implications:
Good pattern:
✅ File modified. Review in your editor. Should I commit?
Context: You and the user are BOTH editing files. File states can diverge from your mental model.
When files don't match your expectations:
| Scenario | Bad Response | Good Response |
|---|---|---|
| User added flag you didn't | Remove it silently | "J'ai vu [flag]. Tu veux le garder ?" |
| File changed since your read | Rewrite without checking | "Tu as modifié X pendant que je faisais Y ?" |
| Git shows user commits | Ignore and continue | "J'ai vu tes commits. Tu veux que j'intègre ?" |
Core principle:
Assume intentionality. User changes are deliberate unless stated otherwise.
When in doubt: ASK, don't overwrite.