Capture concept

Capture Concept

Capture Concept is an Obsidian knowledge distillery. It extracts concepts from cognitive work, decides the capture scenario, calibrates depth, researches gaps, and writes interlinked wiki-grade notes into the user's chosen vault.

Portable means: never assume a machine-specific vault path, folder taxonomy, or private tag schema. This skill is for Obsidian users; require an Obsidian vault.

Files

SKILL_DIR/
├── SKILL.md
├── scripts/init.py
├── scripts/resolve_target.py
├── scripts/vault_ops.py
├── references/
│   ├── scenarios.md
│   ├── enrichment.md
│   ├── graph-and-tags.md
│   └── quality-checklist.md
└── templates/

Preflight

Set SKILL_DIR to this skill directory. Run setup/diagnostics before the first capture:

python3 "$SKILL_DIR/scripts/init.py" --doctor

If not configured, ask for the user's Obsidian vault path, then persist it:

python3 "$SKILL_DIR/scripts/init.py" --vault "/path/to/obsidian-vault" --create-starter-files

Target resolution order:

1. Path explicitly supplied in the prompt.
2. CAPTURE_CONCEPT_TARGET.
3. ~/.capture-concept/config.json.

The target must contain .obsidian/. If the user wants to create a new vault:

python3 "$SKILL_DIR/scripts/init.py" --vault "/path/to/new-vault" --create-vault --create-starter-files

Required:

• Obsidian vault directory with .obsidian/
• Filesystem read/write access
• companion skills/capabilities: obsidian-cli, obsidian-markdown, obsidian-bases

Recommended commands, checked by init.py --doctor:

• obsidian CLI for search, backlinks, properties, and append operations
• rg for fast local vault search
• defuddle for URL capture

Do not auto-install required companion skills or optional commands. If required companion skills are missing, stop setup/capture and tell the user to install them. If optional commands are missing, continue only when the required companion skills provide an equivalent path; otherwise report the limitation.

Use scripts/vault_ops.py for deterministic metadata/log operations when it fits the edit. It refuses to operate outside the configured Obsidian vault.

python3 "$SKILL_DIR/scripts/vault_ops.py" --vault "$TARGET_DIR" add-related "Concepts/note.md" "other-note"
python3 "$SKILL_DIR/scripts/vault_ops.py" --vault "$TARGET_DIR" set-field "Concepts/note.md" updated "YYYY-MM-DD"
python3 "$SKILL_DIR/scripts/vault_ops.py" append-log "$TARGET_DIR" --entry-json '{"concept":"..."}'

Non-Negotiable Workflow

Do not act as a simple summarizer. Every capture must pass through this pipeline. If the user asks for speed, keep the plan concise but still execute the steps.

1. Scenario Detection

Read references/scenarios.md. Determine and remember:

• SCENARIO (S1-S12)
• INPUT_TYPE
• USER_INTENT
• completeness gaps

Ask follow-up questions only when the completeness check says the note would be materially wrong without the missing input.

2. Concept Extraction and Note Type

Use scenario-specific extraction rules from references/scenarios.md.

Also assign note type:

Signal	Type
"What is X" / "How X works" / "X pattern"	concept
"How to do X" / "Step-by-step"	reference
"X vs Y" / "Compare"	comparison
Decision/tradeoff	decision
Debugging postmortem (S5)	reference
Workflow reverse-engineering (S11)	reference
Aha moment (S10)	insight
Meeting crystallization (S8)	reference

For multi-concept scenarios (S1, S2, S4, S7, S9, S12), present the candidate concept list for pruning before bulk writing. Include why each concept is worth capturing and recommended priority.

3. Scope, Research, and Capture Plan

Calibrate reference | working | comprehensive depth using references/scenarios.md. Apply its research gating matrix.

Before writing, create a capture plan with:

• scenario and note type
• selected concepts
• depth and why
• research needed and sources to verify
• existing-note search queries
• intended folder/location
• tag candidates
• backlink candidates and expected relationship
• template(s) to load
• navigation/log updates

For current tools, libraries, APIs, prices, laws, standards, or fast-moving facts, verify with primary/current sources before enrichment.

4. Existing Note Search

Search by title, slug, aliases, and key phrases:

rg -n --glob '*.md' "{concept-name}|{important-alias}" "$TARGET_DIR"
find "$TARGET_DIR" -iname "*{concept-slug}*.md" -not -path "*/.trash/*"

When available:

obsidian search query="{concept-name}" limit=10
obsidian backlinks file="{related-concept}"

If a note exists, update it: add new knowledge, merge tags, update updated:, and preserve unrelated user content. If it does not exist, create it.

5. Location Selection

Reuse the user's vault conventions first. Detect existing domain folders, numbered folders, project folders, MOCs, and _index.md files.

If no convention is clear, use:

Type	Folder
Concepts/references	Concepts/{Domain}/
Decisions	Decisions/{Domain}/
Debugging postmortems	Postmortems/{Domain}/
Meetings	Meetings/{YYYY}/
Session MOCs	_outputs/

If the vault already has a domain taxonomy, follow it. If the note is company-specific or daily-work-specific, prefer the existing work/project area even when the knowledge domain is technical.

6. Tags, Backlinks, and Stubs

Read references/graph-and-tags.md. Run the tag algorithm and backlink scoring algorithm. Do not template-fill tags.

Rules:

• related: contains only verified existing notes.
• Missing but important concepts go into stub_candidates, not related:.
• Bidirectional backlinks are mandatory when a related note is edited safely.
• If a reverse edit cannot be done, report the exact reason.

7. Write the Page

Read references/enrichment.md, then only the needed template:

Scenario	Template
S1/S2/S7 individual concept	templates/concept.md
S3 comparison	templates/comparison.md
S4 concept/tool	templates/concept.md or templates/reference.md
S5 debugging	templates/postmortem.md
S6 decision	templates/decision.md
S8 meeting	templates/meeting.md
S9 feedback pair	templates/feedback-pair.md
S10 aha moment	templates/insight.md
S11 workflow	templates/workflow.md
S12 domain map	templates/domain-map.md plus templates/concept.md
generic how-to	templates/reference.md

Apply depth-based section inclusion. Do not leave empty placeholder sections.

Use provenance-confidence limits:

Provenance	Max Confidence
extracted	high
inferred	medium
ambiguous	low

Conversation captures are usually inferred; article/paper extracts are usually extracted; conflicting or underspecified sources are ambiguous.

8. Graph, Navigation, and Log

Update existing navigation surfaces when present:

• domain _index.md
• domain/project MOC
• wiki-log.md or capture-log.md

If no log exists, create capture-log.md at the target root.

For 3+ concepts, create a session MOC from templates/session-moc.md. For S12, the session MOC is the domain learning map from templates/domain-map.md.

9. Quality Check and Report

Read references/quality-checklist.md and verify all applicable gates.

Report:

• scenario detected
• concepts captured, skipped, and stub candidates
• files created/updated
• target directory and note style
• depth calibration reasoning
• research performed or intentionally skipped
• tags applied and why
• backlinks added with relevance scores
• navigation/log/MOC updates

URL Capture

If the user provides a URL, extract readable Markdown first when possible:

defuddle parse "<URL>" --md -o /tmp/capture-concept-source.md

If defuddle is unavailable, use the best available extraction method and cite that extraction may be lossy. Capture transferable concepts, not article metadata.

Batch Capture

For explicit batches (capture X, Y, Z), process concepts sequentially, cross-link real relationships, and log the batch as one operation.