daily-report

daily-report

Reads today's session traces and cross-references them with the git provider's recent activity to produce a short Markdown report. The output goes to stdout — nothing is posted anywhere. Users who want the report delivered to a chat pipe it into notify (category 5).

When to use

• At the end of a working day, to answer "what did I actually do today?".
• Before a standup, to compile a from-the-trace view of yesterday's activity (run with --date 2026-04-16).
• Any time an auditor wants a reproducible summary of session outcomes against real git-provider events.

Procedure

1. Resolve the sessions directory for the day

Pick the date (default: today in the local timezone) and the scope:

• Repo-local (default): <repo>/.devboy/sessions/<YYYY-MM-DD>/.
• Global (--global): ~/.devboy/sessions/<YYYY-MM-DD>/.

If neither directory exists, exit with an error that tells the user which path was checked.

2. Begin a trace for this report

result=$(devboy trace begin --skill daily-report)
SESSION_DIR=$(echo "$result" | jq -r .session_dir)
SESSION_ID=$(echo "$result" | jq -r .session_id)

This skill is itself traceable — retro runs will see that a daily report ran, how long it took, and whether it succeeded.

3. Walk the per-session subdirectories

The trace subsystem writes each session under <YYYY-MM-DD>/<skill>/<session_id>/, so a single skill can produce several sibling session directories on the same day. Walk every <skill>/<session_id>/ pair under the day:

1. If meta.json is missing, skip the directory — the session is still in flight. Record a single note event listing the skipped sessions so they show up in the report as "in progress".
2. Otherwise, read meta.json and pull:
   • skill, outcome, tool_calls, errors,
   • summary, started_at, ended_at.
3. Aggregate per-skill counts across all that skill's session directories: total runs, success / failure / aborted, total tool-calls, total errors, average duration.

Emit an artifact event per session directory so the retro skill can later find the raw data:

devboy trace event ... --phase artifact \
  --payload "$(jq -nc --arg path "$SESSION_DIR" '{path:$path,kind:"session-dir"}')"

4. Cross-reference with the provider

Two tool calls, each wrapped in a tool_call / tool_result pair:

devboy tools call get_merge_requests \
  '{"state":"merged","limit":50}'
devboy tools call get_issues \
  '{"state":"closed","limit":50}'

Filter the returned lists down to items whose merge / close timestamp falls on the report's date. Neither get_merge_requests nor get_issues has a first-class since parameter in the current schema — do the date filter in the skill, not on the provider side.

Record ok:false on the tool_result if the call fails, but keep going; a daily report is useful even if the tracker is offline.

5. Assemble the Markdown report

Structure:

# Daily report — <YYYY-MM-DD>

## Sessions
- run-and-verify — 8 runs (7 success, 1 failure, avg 4.2s)
- solve-issue — 2 runs (2 success, avg 2m31s)
...

## Merged MRs
- mr#482 — "fix(auth): refresh token rotation"
- mr#485 — "refactor(api): split user service"

## Closed issues
- DEV-411 — "Cannot invite user with plus-sign email"

## Notable failures
- run-and-verify 14:02 — "cargo test" failed on integration::auth

Skip any section that has no entries. Keep the report short — a hundred lines at most.

6. End the trace

devboy trace end \
  --session-dir "$SESSION_DIR" --session-id "$SESSION_ID" \
  --skill daily-report \
  --outcome "$OUTCOME" \
  --summary "<N> sessions, <M> MRs merged, <K> issues closed"

Print the assembled Markdown to stdout and exit.

Success criteria

• The report lists one line per skill that ran today, with accurate counts drawn from meta.json.
• In-flight sessions (no meta.json) are listed under "in progress" rather than silently dropped.
• MRs and issues with a date-stamp outside the target day are not shown.
• The skill never mutates anything — no comments posted, no issues touched.

Guardrails

• Redacted trace payloads stay opaque. If a trace line contains <redacted:...>, carry it into the report verbatim; do not try to reconstruct the original value.
• If both the repo-local and --global trace directories are absent, exit with a non-zero status rather than producing an empty report.

Non-goals

• This skill does not post the report. Use notify from category 5 if delivery is needed.
• It does not create issues, update tickets, or touch git in any way.
• It does not analyse long-term trends — that is retro.