From project-toolkit
Triages raw unstructured input like transcripts and brain dumps into evaluated thread inventories and synthesized gold-found files across extract, evaluate, and synthesize phases.
npx claudepluginhub rjmurillo/ai-agents --plugin project-toolkitThis skill uses the workspace's default tool permissions.
Turn raw, unstructured capture into an evaluated, actionable inventory of threads. Three phases: Extract, Evaluate, Synthesize.
Triages raw unstructured input like transcripts and brain dumps into evaluated thread inventories and synthesized gold-found files across extract-evaluate-synthesize phases.
Extracts deep context from raw sources like transcripts, documents, and sessions. Builds extraction plans, tracks processed items, discovers people, subjects, patterns, and routes to structured bundles.
Processes audio recordings, transcripts, podcasts, lectures into structured Obsidian notes with action items, decisions, glossary. Runs intake interview; suggests agent chaining.
Share bugs, ideas, or general feedback.
Turn raw, unstructured capture into an evaluated, actionable inventory of threads. Three phases: Extract, Evaluate, Synthesize.
| Trigger Phrase | Operation |
|---|---|
pan for gold | Run the full extract -> evaluate -> synthesize procedure |
triage transcript | Build inventory of threads from a raw transcript |
synthesize gold-found | Generate the final gold-found file from evaluations |
Use this skill when:
Use a different skill when:
The skill operates inside a workspace root. The default is ./.panning/ in the current working directory; override via --workspace <path> or the PANNING_WORKSPACE environment variable.
<workspace>/
|-- transcripts/ # Raw input (read-only after capture)
|-- inventories/ # Pass1 + final inventories
|-- evaluations/ # Per-thread evaluation files
`-- gold-found/ # Synthesized gold-found files
The script creates missing subdirectories on demand. It never overwrites an existing file unless --force is passed.
Build a pass1 inventory from a transcript without filtering. Every thread that surfaces is recorded.
Each thread block uses this structure:
## Thread N: <short title>
- **Signal**: high | medium | low
- **Quote**: "<verbatim quote that surfaced this thread>"
- **Context**: <one sentence on where this came from>
- **Initial take**: <one sentence on why this might matter>
The pass1 inventory is provisional. A second pass refines and merges into a final inventory.
For each thread in the final inventory, write an evaluation file. The evaluation captures depth: what evidence supports the thread, what would falsify it, what action it implies.
Evaluations are independent. They can be authored serially or in parallel.
Combine the final inventory and the evaluation files into a gold-found markdown file. The gold-found file groups threads by signal level (High, Medium, Low) and includes a metadata block at the top.
pan.py init --workspace <path>
pan.py validate --inventory <file>
pan.py merge --pass1 <file> --final <file> --output <file>
pan.py synth --inventory <file> --evaluations <dir> --output <file>
The script delegates to inventory.py for parsing and merging, and synthesis.py for gold-found generation.
Signal, Quote, Context, Initial take fields.references/inventory-template.md - thread inventory skeleton.references/gold-found-template.md - gold-found skeleton.