Pinpoint / 定点 is a research-framing skill. It turns broad, vague, or mixed questions into a clearer and more researchable version before deeper analysis starts.
- Claude Code: yes
- Gemini CLI: yes
- Codex-style environments: yes
- Chat apps or web UIs without skill loading: not directly; use the
SKILL.mdcontent as a manual system prompt instead
The portable core of this skill is:
SKILL.mdreferences/
The file agents/openai.yaml is only an environment-specific adapter for the current workspace.
Install pinpoint into your Claude skills directory:
ln -s /absolute/path/to/pinpoint ~/.claude/skills/pinpointOr copy it:
cp -R /absolute/path/to/pinpoint ~/.claude/skills/pinpointThen invoke it in Claude Code with natural language, for example:
Use pinpoint to turn this topic into a researchable question: AI agents in healthcare
Place the pinpoint directory under your Gemini CLI skills directory or configured skills root, then activate it by name:
activate_skill pinpoint
Example prompt after activation:
Turn this topic into 2-3 researchable directions and a short brief: AI agents in healthcare
Notes:
Pinpointdoes not depend on subagents.- Gemini CLI's lack of Claude-style
Tasksupport does not block this skill. - Tool names referenced by shared skill infrastructure may need Gemini CLI equivalents such as
activate_skill.
pinpoint/
├── SKILL.md
├── README.md
├── agents/
│ └── openai.yaml
└── references/
├── output-examples.md
├── phase-2-framework.md
├── phase-3-evidence.md
├── phase-4-counterarguments.md
├── phase-5-scenarios.md
└── phase-6-synthesis.md
Phase 1 only:
- preserve the original question
- diagnose whether it is already researchable
- narrow scope and boundaries where needed
- propose 2-3 candidate research directions when necessary
- recommend one version
- produce a compact research brief
- stop for confirmation
This skill is intentionally front-loaded. Its job is to frame the research well before evidence gathering starts.