Define requirements for an OpenClaw integration
Document a proposed integration's data needs, actions, credentials, and review boundaries.
Use when the what/why is unclear; output is a direction, not a solution.
--- name: discovering description: Clarify goals, scope, and requirements to produce an agreed direction. Use when idea is vague, multiple interpretations exist, scope is undefined, or stakeholders need alignment before technical work begins --- # Discovering Skill Use when the what/why is unclear; output is a direction, not a solution. ## When to Use vs Researching ``` What's unclear? ├── What/Why (goal, scope, requirements) → Discovering └── How (technical approach, implementation) → Researching ``` ## When to Use - Idea is vague or underspecified - Multiple interpretations exist - Need to align on scope and goals - Stakeholders have different expectations Skip if: request is clear and decision-ready (use **researching** skill instead). ## Quick Reference | Phase | Focus | Output | | ---------- | ------------------------------ | --------------- | | Context | Code, docs, recent changes | Current state | | Intent | Purpose, constraints, criteria | Requirements | | Directions | 2-3 approaches + trade-offs | Recommendation | | Confirm | Sign-off | Discovery brief | ## Core Rules - Gather project context first - One question at a time - Prefer multiple choice - High-level directions, not technical solutions ## Flow 1. **Context** — Check code, docs, recent changes 2. **Intent** — Purpose, constraints, success criteria, scope 3. **Directions** — 2-3 high-level approaches with trade-offs; recommend one 4. **Confirm & Output** — Discovery brief: - Goal (1 sentence) - Non-goals - Direction/approach - Open questions (1-3 bullets) Get sign-off before proceeding. ## Common Mistakes | Mistake | Fix | | --------------------------------- | ----------------------------------------- | | Jumping to technical solutions | Stay high-level; how comes later | | Asking multiple questions at once | One question, wait for answer | | Skipping context gathering | Always check existing code/docs first | | No clear recommendation | Always pick one direction and explain why | | Proceeding without sign-off | Get explicit confirmation | ## Delegate - Use **explorer** subagent for a breadth-first repo map and evidence from code/docs/recent history (Context phase) ## After Discovery - Technical uncertainty? → Use **researching** skill - Direction clear? → Use **blueprinting** skill Pipeline: **discovering** → researching → blueprinting → implementing → code-review--- name: discovering description: Clarify goals, scope, and requirements to produce an agreed direction. Use when idea is vague, multiple interpretations exist, scope is undefined, or stakeholders need alignment before technical work begins --- # Discovering Skill Use when the what/why is unclear; output is a direction, not a solution. ## When to Use vs Researching ``` What's unclear? ├── What/Why (goal, scope, requirements) → Discovering └── How (technical approach, implementation) → Researching ``` ## When to Use - Idea is vague or underspecified - Multiple interpretations exist - Need to align on scope and goals - Stakeholders have different expectations Skip if: request is clear and decision-ready (use **researching** skill instead). ## Quick Reference | Phase | Focus | Output | | ---------- | ------------------------------ | --------------- | | Context | Code, docs, recent changes | Current state | | Intent | Purpose, constraints, criteria | Requirements | | Directions | 2-3 approaches + trade-offs | Recommendation | | Confirm | Sign-off | Discovery brief | ## Core Rules - Gather project context first - One question at a time - Prefer multiple choice - High-level directions, not technical solutions ## Flow 1. **Context** — Check code, docs, recent changes 2. **Intent** — Purpose, constraints, success criteria, scope 3. **Directions** — 2-3 high-level approaches with trade-offs; recommend one 4. **Confirm & Output** — Discovery brief: - Goal (1 sentence) - Non-goals - Direction/approach - Open questions (1-3 bullets) Get sign-off before proceeding. ## Common Mistakes | Mistake | Fix | | --------------------------------- | ----------------------------------------- | | Jumping to technical solutions | Stay high-level; how comes later | | Asking multiple questions at once | One question, wait for answer | | Skipping context gathering | Always check existing code/docs first | | No clear recommendation | Always pick one direction and explain why | | Proceeding without sign-off | Get explicit confirmation | ## Delegate - Use **explorer** subagent for a breadth-first repo map and evidence from code/docs/recent history (Context phase) ## After Discovery - Technical uncertainty? → Use **researching** skill - Direction clear? → Use **blueprinting** skill Pipeline: **discovering** → researching → blueprinting → implementing → code-review
Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.
Resource notes updated 2026-10-27 · Original instructions by azat-io