Idea Capturer
Audience: Agents helping users externalize, develop, and connect ideas without losing them.
Goal: Minimize capture friction, then apply the right structure at the right moment — capture, develop, or organize — without over-processing.
The Core Heuristic
At the moment a thought arrives, do not ask "is this worth saving?" — ask "could I forget this?". If yes, capture first, judge later. Judgement at capture time kills the 80% of ideas that only reveal their value in retrospect.
Mode Decision Tree
Pick exactly one mode per interaction. Switching modes mid-flow destroys the capture discipline.
| User signal |
Mode |
What to do |
| "capture", "save this", "idea dump", thought fragment |
Capture |
Minimum friction. Load Quick Capture template. Ask at most one clarifying question. |
| "develop", "flesh out", "what if", half-formed concept |
Develop |
Load Developed Idea template. Ask sharpening questions one at a time. |
| "review", "organize", "connect", "what have I been thinking about" |
Organize |
Surface patterns across existing captures. Propose links, not answers. |
| "brainstorm", "generate ideas about X" |
Brainstorm |
Divergent first (quantity), converge only on explicit request. |
| "synthesize", "what's the pattern across these" |
Synthesize |
Multi-idea input required. Produce one insight, not a list. |
If the user mixes signals, stay in the earliest mode (Capture > Develop > Organize) — you can always develop later; you cannot un-critique a capture.
Methodology Pointers
Claude already knows these frameworks. Invoke by name; do not re-explain in output:
- Zettelkasten — four tiers (fleeting, literature, permanent, project). Tier decides lifespan and review cadence, not how to write the note.
- SCAMPER — for the Develop mode when the user asks "how could this be different?".
- 5 Whys — for the Develop mode when the surface idea is vague ("build a productivity app"); stop at the first answer that names a concrete constraint.
- Divergent then convergent — for Brainstorm mode; never interleave.
See references/templates.md for all capture and development templates. Load the section matching the active mode.
NEVER List
- NEVER batch-process captures more than 24 hours old without re-grounding with the user — stale context leaks into synthesis and produces confident-sounding insights the user never had.
- NEVER link a new idea to more than 3 existing notes without explicit user confirmation — over-linking collapses the idea graph into one homogeneous blob where nothing stands out.
- NEVER critique an idea during Capture mode — even gently ("have you considered..."). The critique is remembered; the idea is lost.
- NEVER ask the user to pick a Zettelkasten tier at capture time. Default to Fleeting. Tier promotion happens during Organize mode, not Capture.
- NEVER convert a capture into a project plan unsolicited. "Save this idea" is not "help me build this".
- NEVER write more than the user wrote, in Capture mode. If the user typed two sentences, your structured capture is two sentences plus metadata — not a page of development.
- NEVER run Synthesize mode on fewer than three related captures — two ideas is a connection, not a synthesis, and the pattern you invent will be noise.
- NEVER tag with more than 5 tags per note. Tag proliferation destroys findability faster than no tags at all.
- NEVER discard an idea because it duplicates an existing note — flag it as a cross-link candidate. Repeated captures of the same thought mean it matters to the user.
- NEVER use the Develop mode's sharpening questions as an interrogation — ask one, wait, respond to the answer, then ask the next. Batch questions fail.
Idea Types and Development Focus
| Type |
Signal |
Development pivot |
| Project |
"build", "make", "create" |
Feasibility + narrowest-wedge user |
| Question |
"why", "how come", "I wonder" |
Research direction, not answer |
| Insight |
"I realized", "turns out" |
Connection to prior notes, application |
| Observation |
"I noticed", "pattern I see" |
Pattern-naming, not interpretation |
| Spark |
"reminds me of", "what if" |
Capture now, explore only on request |
Quick Reference
| Action |
Trigger |
| Quick capture |
"capture idea", "save this" |
| Develop |
"develop idea about X" |
| Brainstorm |
"brainstorm X" |
| Review captures |
"review my ideas" |
| Find related |
"find ideas about X" |
| Synthesize |
"synthesize these" |
Integration Points
- Journal/reflection tools surface ideas worth capturing.
- Obsidian or second-brain vaults are the storage layer; this skill is the ingestion discipline on top.
concept-cartographer can visualize the idea graph once enough captures accumulate.
1---2name: idea-capturer-23description: Structured system for capturing fleeting thoughts, developing raw ideas into actionable concepts, and organizing creative thinking. Use when the user wants to capture an idea, brain-dump, develop or explore a half-formed concept, run a brainstorming session, organize scattered notes, synthesize ideas, or build a Zettelkasten/second-brain workflow. Trigger phrases: "capture idea", "save this idea", "idea dump", "brain dump", "develop idea", "brainstorm", "connect ideas", "review my ideas", "tag idea", "find ideas about".4---56# Idea Capturer78**Audience:** Agents helping users externalize, develop, and connect ideas without losing them.9**Goal:** Minimize capture friction, then apply the right structure at the right moment — capture, develop, or organize — without over-processing.1011## The Core Heuristic1213At the moment a thought arrives, **do not ask "is this worth saving?"** — ask **"could I forget this?"**. If yes, capture first, judge later. Judgement at capture time kills the 80% of ideas that only reveal their value in retrospect.1415## Mode Decision Tree1617Pick exactly one mode per interaction. Switching modes mid-flow destroys the capture discipline.1819| User signal | Mode | What to do |20|---|---|---|21| "capture", "save this", "idea dump", thought fragment | **Capture** | Minimum friction. Load Quick Capture template. Ask at most one clarifying question. |22| "develop", "flesh out", "what if", half-formed concept | **Develop** | Load Developed Idea template. Ask sharpening questions one at a time. |23| "review", "organize", "connect", "what have I been thinking about" | **Organize** | Surface patterns across existing captures. Propose links, not answers. |24| "brainstorm", "generate ideas about X" | **Brainstorm** | Divergent first (quantity), converge only on explicit request. |25| "synthesize", "what's the pattern across these" | **Synthesize** | Multi-idea input required. Produce one insight, not a list. |2627If the user mixes signals, stay in the **earliest** mode (Capture > Develop > Organize) — you can always develop later; you cannot un-critique a capture.2829## Methodology Pointers3031Claude already knows these frameworks. Invoke by name; do not re-explain in output:3233- **Zettelkasten** — four tiers (fleeting, literature, permanent, project). Tier decides *lifespan and review cadence*, not *how to write the note*.34- **SCAMPER** — for the Develop mode when the user asks "how could this be different?".35- **5 Whys** — for the Develop mode when the surface idea is vague ("build a productivity app"); stop at the first answer that names a concrete constraint.36- **Divergent then convergent** — for Brainstorm mode; never interleave.3738See `references/templates.md` for all capture and development templates. Load the section matching the active mode.3940## NEVER List41421. **NEVER** batch-process captures more than 24 hours old without re-grounding with the user — stale context leaks into synthesis and produces confident-sounding insights the user never had.432. **NEVER** link a new idea to more than 3 existing notes without explicit user confirmation — over-linking collapses the idea graph into one homogeneous blob where nothing stands out.443. **NEVER** critique an idea during Capture mode — even gently ("have you considered..."). The critique is remembered; the idea is lost.454. **NEVER** ask the user to pick a Zettelkasten tier at capture time. Default to Fleeting. Tier promotion happens during Organize mode, not Capture.465. **NEVER** convert a capture into a project plan unsolicited. "Save this idea" is not "help me build this".476. **NEVER** write more than the user wrote, in Capture mode. If the user typed two sentences, your structured capture is two sentences plus metadata — not a page of development.487. **NEVER** run Synthesize mode on fewer than three related captures — two ideas is a connection, not a synthesis, and the pattern you invent will be noise.498. **NEVER** tag with more than 5 tags per note. Tag proliferation destroys findability faster than no tags at all.509. **NEVER** discard an idea because it duplicates an existing note — flag it as a cross-link candidate. Repeated captures of the same thought mean it matters to the user.5110. **NEVER** use the Develop mode's sharpening questions as an interrogation — ask one, wait, respond to the answer, then ask the next. Batch questions fail.5253## Idea Types and Development Focus5455| Type | Signal | Development pivot |56|---|---|---|57| Project | "build", "make", "create" | Feasibility + narrowest-wedge user |58| Question | "why", "how come", "I wonder" | Research direction, not answer |59| Insight | "I realized", "turns out" | Connection to prior notes, application |60| Observation | "I noticed", "pattern I see" | Pattern-naming, not interpretation |61| Spark | "reminds me of", "what if" | Capture *now*, explore only on request |6263## Quick Reference6465| Action | Trigger |66|---|---|67| Quick capture | "capture idea", "save this" |68| Develop | "develop idea about X" |69| Brainstorm | "brainstorm X" |70| Review captures | "review my ideas" |71| Find related | "find ideas about X" |72| Synthesize | "synthesize these" |7374## Integration Points7576- Journal/reflection tools surface ideas worth capturing.77- Obsidian or second-brain vaults are the storage layer; this skill is the ingestion discipline on top.78- `concept-cartographer` can visualize the idea graph once enough captures accumulate.
praxstack