sra

name	sra
description	Run the Stock Research Agent equity-research pipeline for a ticker via a Claude Code workflow. Use when asked to "run SRA", "research <TICKER>", or "generate an equity research report".

Stock Research Agent (workflow runner)

Runs the full equity-research DAG (dags/sra.yaml) as an in-session workflow so all model work uses the Max subscription, with the /workflows progress UI and sidecar-based resume.

Usage

/sra TICKER [--date YYYYMMDD] [--length short|standard|long]

Steps (in order)

1. Collect custom prompts. Ask: "Any custom investigation prompts for ? (specific risks/segments — or 'none')." Collect as a list (empty if 'none').

2. Run prep (no model calls) via Bash from the repo root:

   uv run python prepare_run.py <TICKER> [--date ...] [--length ...] --ask "<p1>" "<p2>" ...
   

   stdout is the resolved DAG JSON. Note workdir.

3. Build the workflow script. The resolved DAG (~190 KB with prompts inline) is too large to pass via args (string limit) or an agent echo (Read caps at ~25K tokens), so embed it as a literal in the script. prepare_run.py already wrote it to <workdir>/.sra/dag.json. Assemble <workdir>/sra_workflow.js via Bash (concatenation — do NOT hand-author the DAG):

   WF=<workdir>/sra_workflow.js
   cat > "$WF" <<'META'
   export const meta = {
     name: 'sra-<TICKER>',
     description: 'Equity research pipeline for <TICKER>',
     phases: [ {title:'Gather'},{title:'Index'},{title:'Research'},
               {title:'Index research'},{title:'Write'},{title:'Assemble'} ],
   }
   META
   printf 'const __SRA_DAG__ = ' >> "$WF"
   cat <workdir>/.sra/dag.json   >> "$WF"
   printf ';\n'                  >> "$WF"
   cat workflows/sra_interpreter.body.js >> "$WF"
   

   The body reads the embedded __SRA_DAG__. (Total script stays < 512 KB.) Validate with node --check against an async-wrapped copy if desired.

4. Launch. Call the Workflow tool with scriptPath: <workdir>/sra_workflow.js and no args (the DAG is embedded). The interpreter uses __SRA_DAG__.

5. Report the returned {completed, incomplete, artifacts} and the final report location under <workdir>/artifacts/.

Resume

Re-run the same command + --date. prepare_run.py reads .sra/tasks/*.json sidecars; the interpreter skips completed tasks and rehydrates their artifacts/variables.