Summarize Period

Files (1)

• SKILL.md

SKILL.md (first lines)

## Input
For weekly: all daily profiles + previous weekly. For monthly: all weekly summaries + previous monthly.

## Output (JSON)
```json
{
  "themes": [{ "label": "auth migration", "effort_share": 0.6, "progress": "Completed token refresh." }],
  "accomplishments": ["Merged PR #1234"],
  "stuck": [{ "description": "ARM CI pipeline flaky", "days_or_weeks": 4 }],
  "focus_next": ["Finish auth rollout"]
}
```

## Instructions
1. Merge focus_areas across input profiles — weight by confidence and frequency.
2. Compute effort_share per theme (sum to ~1.0).
3. Extract accomplishments: merged PRs, resolved threads, milestones. Deduplicate.
4. Extract stuck items: unfinished on 2+ days (weekly) or 2+ weeks (monthly).
5. Suggest focus_next from momentum + carryover + stuck.
6. If previous summary provided, absorb unresolved items into absorbed_from_previous.
7. **Weekly only — aggregate brief quality:** Scan the `feedback` field across all input daily profiles. For each profile with a `feedback` entry, tally `rated` (count), `positive` (count where `sentiment == "positive"`), `negative` (count where `sentiment == "negative"`), and collect all distinct negative `detail` strings into `themes`. Emit as `brief_quality: { rated, positive, negative, themes }`. If no daily profiles had a `feedback` field, omit `brief_quality`.
8. **Monthly only — roll up brief quality:** Sum `rated`, `positive`, and `negative` across all input weekly `brief_quality` fields. Merge `themes` arrays and deduplicate. Emit as `brief_quality`. If no weekly summaries had `brief_quality`, omit the field.
9. Preserve the input periods' `impact_summaries` as evidence for **dayarc-write-impact-summary**. Do not flatten
   them into generic accomplishment bullets or invent impact here; the shared impact skill performs that synthesis.

From YuiZhou/dayarc-agent, MIT.

More from YuiZhou/dayarc-agent