kestral-end-day-review

End Day Review

Produce an evidence-backed close-out for today and a practical priority list for tomorrow. Gather current context first, then ask before writing anything back to Kestral or local project files, except .kestral/preferences.md memory updates covered by the User preferences rules.

Prerequisites

The Kestral MCP server must be in this session (/mcp). Call whoami first — if it fails, ask the user to reconnect or authenticate the Kestral MCP server in their app's MCP settings (a browser should open for sign-in). Do not call other tools until authenticated. Calendar access is optional, but use it when tomorrow prioritization depends on schedule realism.

Human-readable references

Keep Kestral IDs internal unless the user asks for them. In user-facing output:

• Tasks: show slug - title when a slug is available, linked with url when the host can render links.
• Projects, documents, feedback, customers, tags, statuses, and other Kestral entities: show the readable name/title/label first, linked with url when the host can render links.
• People and actors: show display names; if unresolved, write Unknown member (id: <rawId>).
• Unknown non-member entities: write Unknown <entity type> (id: <rawId>).
• Approval tables and write-back plans must put the human-readable label first. Raw URLs, machine IDs, source IDs, and bare slugs belong only in secondary metadata when useful.
• Use existing display fields first; do extra lookups only for entities that matter to the answer.

Entrypoint

Expected invocations include:

• /kestral:end-day-review
• $kestral-end-day-review
• "What got done today and what didn't?"
• "Review today's sessions and Kestral updates, then tell me what to prioritize tomorrow."

Data gathering order

Default order:

1. Get today's date, timezone, and relevant agent session history for the local day. Session stores are agent-specific: for example, Codex uses ~/.codex/sessions, Claude Code uses ~/.claude/projects, and other agents may use their own local or hosted transcript stores.
2. Read .kestral/preferences.md by checking the current workspace folder and then parent folders. Use the first match as the source for durable user close-out and tomorrow-planning preferences, not as task or project state.
3. Fetch the latest Kestral daily brief with get_daily_brief.
4. Search Kestral tasks, projects, feedback, knowledge, and document chunks for today, especially Project Brain or named projects from the brief and agent session history.
5. Read relevant Kestral entities directly when search results identify exact project, task, document, or project brain IDs.
6. Check remote GitHub context when available: current branch commits, open or recently updated PRs, linked PRs, and commit or PR references in relevant task comments.
7. Check connected MCPs/apps that are directly relevant to tomorrow prioritization, especially Calendar when the user's day has scheduling constraints.
8. Read local project files only when they are relevant to surfaced projects or updates, such as a matching overview.md or repo plan file.

Do not treat one source as authoritative when it conflicts with fresher live state. Prefer exact Kestral entity state over generated summaries, and call out stale or unavailable sources.

Kestral searches to run when close-out needs verification:

• tasks updated today <project name>
• urgent open tasks in <project name>
• blocked tasks <project name>
• documents updated today <project name>
• project-level search_content (type: "knowledge") for current blockers, next steps, and progress

Calendar searches should use explicit local-day RFC3339 bounds for tomorrow. If calendar access is missing or empty, do not infer a free day; state the gap.

Run independent reads in parallel whenever the host supports it: local preferences, daily brief, agent session searches, remote GitHub reads, broad Kestral searches, and tomorrow's calendar query do not need to block one another. Do not parallelize dependent lookups; fetch exact entities only after search results identify the relevant IDs.

User preferences

Treat .kestral/preferences.md as a local memory file for durable review and prioritization preferences. Find it by walking upward from the current workspace folder and using the first match.

• Read the file before summarizing today or ranking tomorrow. If it does not exist, continue without it and mention the absence only when a preference write is proposed.
• Apply relevant saved preferences when recommending tomorrow priorities, such as preferred close-out format, decision style, focus-hour defaults, recurring projects to check, communication cadence, and work the user consistently wants avoided.
• Capture durable preference signals from the user's constraints, corrections, repeated edits, and stated likes/dislikes. Do not require the user to explicitly say "remember", "note", "save", or "prefer".
• Do not treat one-day constraints or today's mood as durable preferences. Save only stable work-style, close-out, prioritization, scheduling, or write-back preferences that are likely to apply across future end-day reviews.
• Update .kestral/preferences.md silently when a durable preference is clear. This is local memory maintenance, not a Kestral write-back.
• Ask before writing preferences only when the signal is ambiguous, appears one-off, conflicts with existing memory, or may include sensitive personal or meeting-specific content.
• Create .kestral/ and preferences.md in the current workspace folder if no parent preference file exists and a preference write is needed. Keep the file short, in Markdown, and update existing bullets instead of appending duplicates.
• Do not store credentials, private personal details, or sensitive meeting content in preferences.

Workflow

1. Gather today's agent history

Build a compact evidence list before summarizing:

• User requests, agent actions, session transcripts, and relevant session data from today's agent-specific session history.
• Kestral daily brief highlights and timestamp.
• Kestral project/task/document changes from today.
• Remote GitHub PRs, current branch commits, linked PRs, and commit references when available.
• Relevant Project Brain/project brain updates, proposed changes, blockers, and decisions.
• Calendar constraints for tomorrow if prioritization depends on available time.
• Saved close-out or prioritization preferences that materially affect the review.
• Local overview.md or project docs that are relevant to surfaced projects or updates.

Keep raw session review targeted. Search for today's user messages, final answers, tool calls, project names, task IDs, PR URLs, commit SHAs, branch names, and write-back actions instead of reading every token linearly. If transcript search or session context is unavailable, state the gap and continue from live Kestral, git, GitHub, and available local project evidence.

2. Reconcile done vs not done

Separate confirmed outcomes from attempted or pending work:

• Done: shipped changes, created/updated tasks, documents, PRs, commits, decisions made, bugs fixed, and verified outcomes.
• Not done: explicit unfinished checklist items, blocked work, failed verification, deferred tasks, pending reviews, stale open questions, and work that was discussed but not executed.
• Unclear: items where evidence is missing or conflicting. Say what would verify them.

Use concrete project/task/document names and dates. Avoid vague claims like "made progress" unless the specific artifact is named.

3. Find Project Brain updates

For each relevant project:

• Identify the Kestral project and, when available, its project brain or current generated knowledge.
• Summarize what changed today in goals, blockers, open decisions, next steps, and current status.
• Flag mismatches between live task/project state and generated Project Brain knowledge.
• Identify whether the right write-back is a task mutation, task creation, document update, comment, or overview.md maintenance.

If generated Project Brain knowledge appears stale, say that explicitly and base recommendations on live entities.

4. Recommend tomorrow priorities

Rank tomorrow's work by impact, urgency, unblock value, and calendar realism:

1. Must-win outcome for tomorrow.
2. Two or three next priorities.
3. Quick wins or communications that reduce risk.
4. Deferred work with a reason.

If Calendar is available, fit priorities around known meetings and preserve buffer. If Calendar is unavailable, label the plan as task-priority-only and ask for fixed commitments if needed.

5. Ask before write-back

Never write back automatically. Present a specific write-back plan and wait for user approval.

Valid write-backs:

• Create Kestral tasks for newly discovered follow-ups.
• Update Kestral tasks for status, priority, due date, assignee, or description changes.
• Comment on Kestral tasks with close-out notes.
• Create or update Kestral documents with the daily review.
• Update an existing local overview.md in the relevant project folder.
• Create overview.md if the target project folder has no overview file and the folder is a local draft/project workspace.

For each proposed write, include target, action, and exact content summary. For task changes, include priority/status/due date values. For document or overview.md edits, include the section names that will be changed.

After approval, apply only the approved writes. Return links for Kestral mutations and file paths for local edits.

Preference-memory updates to .kestral/preferences.md are separate from Kestral/local write-backs. Apply them silently under the User preferences rules when durable preferences are clear; mention the local file path only if useful.

Output shape

Use this structure unless the user asks for something else:

## Done Today
- ...

## Not Done
- ...

## Project Brain / Project Updates
- ...

## Tomorrow Priorities
1. ...
2. ...
3. ...

## Recommended Write-Backs
- [human-readable target] [action] - [summary]

Approve these write-backs?

If the user already approved a specific write-back plan, replace the final question with a concise result list.

Quality bar

• Cite evidence with Kestral links, local file paths, task slugs, document titles, PR URLs, or session filenames when possible.
• Keep raw Kestral IDs internal to search and lookup steps. In final summaries and write-back plans, use slug - title for tasks and readable names/titles for projects, documents, feedback, customers, statuses, tags, and members.
• Include data gaps instead of hiding them.
• Do not mutate Kestral, Calendar, GitHub, or local files without approval, except .kestral/preferences.md memory updates covered by the User preferences rules.
• Preserve existing overview.md structure when updating local project folders.