What Have We Done Today

Overview

A manual, on-demand daily recap for any Hermes installation. It scans three local surfaces for today's activity — sessions (default profile plus every named profile, all sources), kanban boards (done today, running, blocked, todo/ready), and cron stores (jobs that ran, errored, are running now, or are due today) — then writes a plain append-friendly daily markdown log the user can read to remember what happened and what is still open.

Stdlib Python only. Read-only on all databases (mode=ro); the only write is the daily markdown file in the recap directory. It is intentionally NOT a cron job: it runs when a human actually wants the recap (tired, stoned, context-switch heavy, or closing out the day).

When to Use

• "What have we done today?" / "Recap today" / "What did I get up to today?"
• The user lost track of a busy multi-project day and wants the threads pulled together.
• Before shutting down for the night: capture progress + leftovers.

Counter-triggers — do NOT load this skill when:

• The user asks about a specific past session or topic (use session search).
• The user wants a recurring scheduled recap (this skill is manual-only).
• The user wants to mutate sessions, kanban cards, or cron jobs.

Workflow

1. Scan all three surfaces. Invoke through the terminal tool: python3 ~/.hermes/skills/productivity/what-have-we-done-today/scripts/today_sessions.py --excerpts Output: sessions grouped by profile (time, source, title, counts, goal/outcome hints), then kanban per board (done today, ▶ running, ✋ blocked, todo/ready counts), then cron (ran today with ❌ on errors, ▶ running now, ⏳ due today after an error).
2. Separate the signals. User-facing session sources are desktop, cli, telegram. kanban sessions are worker lanes (generic titles — outcome comes from the excerpt, not the title). Cron ❌/⏳ rows are real breakage candidates; running executions are work happening right now.
3. Drill where it matters. session_search reads only the current profile's DB; use it for ambiguous or high-stakes default-profile sessions. For sessions from profiles/*/state.db, the scanner's excerpts are the cheap evidence.
4. Split Done vs Open. Done = a verified outcome (session excerpt, kanban done-today card, cron ok run). Open = kanban ▶/✋/todo cards, cron ❌/⏳ rows, sessions with no resolution. Never infer completion from a session merely existing.
5. Pull open items (optional). --ledger FILE lists non-completed items from a task ledger JSON: {"items": [{"id", "title", "status", "next_action", "updated_at"}]}. Statuses other than completed / cancelled count as open.
6. Write the log. Target $RECAP_DIR/YYYY-MM-DD.md (default ~/.hermes/daily-recaps/); create the directory if missing. Create the file with a # YYYY-MM-DD heading if absent; otherwise append a ## Daily recap — HH:MM section with Done / In flight / Needs attention bullets. Re-runs the same day append new sections.
7. Brief the user. Compact and ranked: ✅ Done today, 🟡 Still open, ⛔ Needs you, Next pick. Include @session:default/<id> links for anything they might reopen.

Common Pitfalls

• ~/.hermes/sessions.db may be an empty legacy file — the real session store is ~/.hermes/state.db plus profiles/*/state.db. A scan of the wrong file reports a false quiet day.
• started_at is epoch seconds; compare against local midnight (the scanner does this in Python). SQLite's 'start of day' modifier is UTC midnight and silently drops the first hours of the local day.
• Sessions that straddle midnight still count: the filter is started_at OR last_activity_at ≥ local midnight.
• Each kanban board is its own DB: root ~/.hermes/kanban.db holds only the default board; named boards live at ~/.hermes/kanban/boards/<slug>/kanban.db (_archived/ is skipped).
• no_agent cron jobs leave no session at all — jobs.json last_run_at/last_status and executions.db are their only trace. A quiet sessions list does not mean a quiet day.
• Executions use ISO-text timestamps; sessions use epoch floats. Parse each surface with its own type.
• The recap log is append-friendly by design — never overwrite an existing day file or the earlier recap is destroyed.
• The scanner never writes: every DB opens mode=ro with a busy timeout, so a live Hermes holding the default DB still scans fine.
• Don't install this as a cron — the manual trigger is the design.

Verification Checklist

• Scanner ran with exit code 0 and reported sessions, kanban, and cron.
• The daily log exists and contains a recap section for today: grep -n "Daily recap" ~/.hermes/daily-recaps/$(date +%F).md
• Re-running the scanner returns the same inventory.
• Every open item in the briefing traces to an excerpt, kanban card, or ledger row read during this run.