| name | flex:digest |
| description | Workstream-oriented standup of recent coding-agent activity. Runs focused SQL over a sessions cell (inventory, file hotspots, delegations, open loops), then synthesizes the combined payload into a coworker-style digest. |
| user-invocable | true |
| argument-hint | optional cell + window, e.g. 'claude_code today', 'codex 3d' |
flex:digest
The programmatic version of @digest: instead of one preset result, run four focused queries and synthesize the combined payload. Works against any coding-agent sessions cell (claude_code, codex, ...).
Data Flow
queries/digest/*.sql
-> scripts/flex-query.sh <cell>
-> model synthesis
Execution
From this skill directory (default cell claude_code; substitute any sessions cell):
cat queries/digest/inventory.sql | scripts/flex-query.sh claude_code
cat queries/digest/files.sql | scripts/flex-query.sh claude_code
cat queries/digest/delegations.sql | scripts/flex-query.sh claude_code
cat queries/digest/open-loops.sql | scripts/flex-query.sh claude_code
The four phases:
- Inventory — recent non-agent sessions: community labels, centrality, hub/bridge flags, fingerprints, project counts.
- Files — last-day Read/Edit/Write hotspots, grouped by file identity when available.
- Delegations — parent and child agent relationships.
- Open loops — overnight sessions and large sessions that may need pickup.
Extra semantic assets (not part of the standard run, useful for deeper passes): user-voice.sql, user-voice-tight.sql, secondary-work.sql, workstream-clusters.sql.
Synthesis Instructions
After the four payloads are in, DO NOT run additional queries. Synthesize:
- Write like a coworker giving a casual standup, not a changelog. The reader scans in 20 seconds.
- Group by workstream (community labels + fingerprints carry the substance — use them to say what actually happened, not that work "continued").
- Surface file hotspots only when they sharpen a workstream's story.
- End with open loops: what looks unfinished, what may need pickup today, and any delegation chains left hanging.
- Preserve uncertainty instead of forcing every session into a taxonomy.
Rules
- Start from evidence, not vibes.
- Keep SQL in
queries/, mechanics in scripts/, synthesis rules here.
- Treat scores/centrality as ordering signals within one digest, never as absolute values.
