Daily Synth

Daily Synth

Use this skill for daily task aggregation and executive focus.

Required Preflight

Before synthesizing, run the critical commitment preflight:

python3 system/scripts/critical_commitment_refresh.py plan --mode day --json
python3 system/scripts/critical_commitment_refresh.py rank --mode day --json

If the plan reports should_pause_for_user: true, stop and show the user:

• What integration failed.
• Why the skipped source puts commitments at risk.
• The recommended fix.
• Safe options: reconnect/configure, paste/export for this run, skip once, or disable from defaults.

Do not continue with degraded source coverage until the user explicitly chooses a path.

Source Policy

• Use manifest-backed named read-only source windows by default for Slack, Outlook, Calendar, Teams, transcripts, Quill, Granola, Obsidian, Atlassian, and agent memory.
• Enabled, healthy TWG may supplement the named Atlassian window with one bounded personal work rollup under .agent/rules/TWG_READ_ONLY.md. Rovo/native reads remain primary for exact Jira or Confluence evidence.
• Backward source windows default to the last 5 business days and may shorten only when the chat/source manifest or command-run manifest shows a successful prior run for the same source/window. Calendar includes the last 5 business days of changes plus forward lookahead for upcoming active-workstream gates.
• Treat third-party systems as read-only. Never send, draft, reply, react, schedule, create, assign, transition, comment, upload, patch, delete, or move third-party state unless the user explicitly confirms that exact mutation in the current turn.
• Use parallel agents for independent source reads and synthesis passes when the runtime supports them.
• Keep shared writes serialized: Markdown task updates, source manifests, generated indexes, and final consolidation.
• Stay within the 90-second wall-clock budget where possible. Prefer visible partial source gaps over silent skips.

Synthesis Rules

• Direct manager, skip-level, and executive callouts outrank ordinary stale work.
• External partner/customer/end-user commitments with agreed dates outrank internal undated work.
• Completed outcomes stay visible with completion date/source.
• Implied completions become confirmation questions.
• Group output by workstream title first, never by Task Master ID.
• User-facing task and workstream descriptions must use display_title plus source provenance, not hard IDs. Keep internal and source IDs only in local links, metadata, or an Agent refs line.
• Evidence lines use this shape: [Task/workstream title]; started from [initial source label] on [date]; latest progress from [latest source label] on [date].

Output Shape

### Workstream Title
- Latest outcome: [succinct result, decision, or state]
  - Evidence: [task/workstream title; started from initial source on date; latest progress from latest source on date]
- Completed: [done item or "None newly confirmed"]
  - Completed: [date/source]
- Open items: [count or short list]
  - [Owner] - [action] by [date/gate], from [source]
- Recommended next 3:
  - [Action 1]
  - [Action 2]
  - [Action 3]