Read memory/ideas/ and report the portfolio. Do not run any analysis.
Filter: $ARGUMENTS
For every idea directory, read idea.md frontmatter and scores.json if present,
then produce a table ordered by score descending:
| Idea | Model | Stage | Score | Verdict | Status | Missing |
|---|
- Stage is
scoring_stage: candidate-quick-score, fast-validation or full-validation. - Verdict is the verdict for a full validation, or the rank label for a quick score. Never show a verdict for a quick score.
- Missing is the count of dimension files not yet written, from
missing_inputs.
Then call out, in prose:
- Ideas with dimension files but no
decision_memo.md- started and abandoned. - Ideas still at
status: candidatewhose scores came from a capped competition prior, so their ranking is provisional. - Any idea whose
superseded_bypoints at a pivot, with where it went.
Finish with one recommendation: which idea deserves the next hour, and why. If a single slug was given, skip the table and give the full picture for that one idea.