analyze

Analyze

The main entry point. Runs a portfolio of decision frameworks in parallel against a single decision and produces structured per-framework outputs plus a synthesis.

Resolve config

DATA_ROOT="${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/decision-evaluation-framework"
CONFIG="$DATA_ROOT/config.json"

If the config doesn't exist, tell the user to run /decision:onboard first.

Inputs

The user invokes this with:

• An existing decision file path, OR
• A description of the decision (in which case run /decision:new-decision first to scaffold a file, then proceed).

Optional flags / arguments:

• --bundle=<name> — use a named bundle from config.
• --frameworks=cost-benefit,pre-mortem,... — explicit framework list.
• (none) — use default_frameworks from config.

Procedure

1. Load the decision file. Confirm it has the expected sections (Context, Options, Constraints, Stakeholders, etc.). If it's incomplete, offer to complete it interactively before continuing.

2. Resolve framework list. Bundle name → list, explicit list, or default. Validate every name corresponds to a file in the plugin's frameworks/ directory.

3. Create the analysis workspace at <output_root>/<YYYY-MM-DD>-<slug>/:

   <slug>/
     decision.md              # copy of the input decision
     fragments/               # per-framework outputs land here
     synthesis.md             # written by /decision:synthesize
     analysis.pdf             # written by /decision:render-pdf
   

4. Fan out IN PARALLEL. Launch N subagents in a single message with N Task tool calls (subagent_type: general-purpose). Each subagent receives:

   Apply the <framework-name> framework to the following decision.
   
   Decision file: <absolute path to decision.md>
   Framework guide: <absolute path to frameworks/<framework-name>.md>
   
   Read the decision fully. Read the framework guide fully. Apply the framework
   exactly as the guide specifies. Write your analysis to:
       <absolute path to fragments/<framework-name>.md>
   
   Use the output structure the guide prescribes. Conclude with a 0-100 score
   per the guide's scoring instruction. Do not invent additional frameworks.
   

   The parallel fan-out is mandatory — sequential execution is much slower and produces less independent perspectives.

5. Wait for all subagents to finish. Verify each fragment file was written. Re-run any that failed.

6. Synthesize. Invoke /decision:synthesize <slug-folder> to produce the cross-framework synthesis (or run that skill's logic inline).

7. Offer to render — ask the user if they want a PDF (/decision:render-pdf) and/or upload to Drive (/decision:export).

8. Report the workspace path and the synthesis verdict.

Notes

• Decision-file format is documented in /decision:new-decision.
• If the user passes --frameworks with an unknown name, list available frameworks and ask.
• Don't pre-bias the synthesis with your own opinion — the synthesis skill reads only the fragments.