reporting-work-activity

Files (28)

• SKILL.md
• openai.yaml · agents/
• .gitignore · assets/activity-report-runtime/
• README.md · assets/activity-report-runtime/
• agent_session_evidence.py · assets/activity-report-runtime/
• bootstrap_tracked_sources.py · assets/activity-report-runtime/
• build_report.py · assets/activity-report-runtime/
• clear_runtime_cache.py · assets/activity-report-runtime/
• .gitkeep · assets/activity-report-runtime/data/
• export_single_page_pdf.py · assets/activity-report-runtime/
• generate_report.py · assets/activity-report-runtime/
• mise.toml · assets/activity-report-runtime/
• muted_slack_channels.json · assets/activity-report-runtime/
• populate_data.py · assets/activity-report-runtime/
• pyproject.toml · assets/activity-report-runtime/
• report_config.py · assets/activity-report-runtime/
• report_design.py · assets/activity-report-runtime/
• report_server.py · assets/activity-report-runtime/
• report_sources.py · assets/activity-report-runtime/
• test_agent_evidence.py · assets/activity-report-runtime/tests/
• test_smoke.py · assets/activity-report-runtime/tests/
• tracked_sources.template.json · assets/activity-report-runtime/
• uv.lock · assets/activity-report-runtime/
• verify_report.py · assets/activity-report-runtime/
• agent-session-evidence.md · references/
• personal-vs-org-reports.md · references/
• runtime-contract.md · references/
• visual-design-contract.md · references/

SKILL.md (first lines)

# Activity Report Refresh

Refresh the bundled report runtime in `assets/activity-report-runtime`. Use the
local tools there for GitHub and Linear, then fill the Slack and Notion gaps
with Codex connectors and write the resulting snapshots under `data/`.

Do not request or run peer review for this workflow, including through the
`reviewing-complex-work` skill.

## Scope first

Decide the report scope before refreshing or rewriting conclusions.

- Personal report: use this when the user says "me", "my activity", "just me",
  "about me", "make me look good", or corrects an org-wide report as too broad.
  Read `references/personal-vs-org-reports.md` before changing content.
- Full org report: use this only when the user asks for Convergint,
  organization, team, workspace, department, or broad weekly activity.
- If the wording is ambiguous, ask a short scope question before doing connector
  work. Do not default from a prior run.

For personal reports, GitHub and Linear should be filtered to Chad's own
activity first. Slack, Notion, and Datadog should support the story instead of
becoming broad digest sections.

Include Chad's Codex, Claude, and Cursor sessions in every personal evidence
pass. Read `references/agent-session-evidence.md` for discovery, timestamp
filtering, corroboration, attribution, and exploration rules. Use sessions
primarily to support completed work. They can also surface significant topics
Chad explored without implying that a proposal, draft, or conversation shipped.

## Lock the reporting window

Resolve the window before clearing cache, fetching data, or reading connector
evidence.

1. Run `date '+%Y-%m-%d %H:%M:%S %Z'`. Do not infer today's date from model
   knowledge or from the previous report.
2. For "my weekly report," "last week's work activity," or another request for
   the latest report, default to the rolling seven days ending today in the
   user's local timezone. Use a previous Monday-through-Sunday week only when
   the user explicitly asks for the previous calendar week.
3. Set one explicit date and timezone for the entire run:

   ```bash
   export REPORT_DATE="$(date +%F)"
   export REPORT_TIMEZONE="America/Chicago"
   ```

4. State the exact inclusive window in a progress update before fetching. The
   runtime derives the start as six days before `REPORT_DATE`.
5. Use the same exported values for `mise run fetch`, `mise run report`, and
   `mise run verify-report`. Never silently repin the report to a prior Sunday.

For APIs with timestamp filters, translate the inclusive local-date window to
`start <= timestamp < day_after_end`. Slack searches should use the equivalent
exact `after:` and `before:` dates.

## Quick start

Copy this checklist and track progress:

```text
Activity report refresh progress:
- [ ] Confirm the bundled runtime path and current local state
- [ ] Decide and record scope: personal report or full org report
- [ ] Look up today's local date and state the exact report window
- [ ] Export one `REPORT_DATE` and `REPORT_TIMEZONE` for the whole run
- [ ] If redesigning, capture the baseline and write a named design brief
- [ ] Seed `tracked_sources.json` if it is missing
- [ ] Clear generated cache unless the user explicitly wants to keep it
- [ ] Run `mise run fetch`
- [ ] Refresh seed Slack and Notion sources
- [ ] Run the bounded discovery pass for Slack and Notion
- [ ] Write `data/slack_channels.json`
- [ ] Write `data/notion_pages.json`
- [ ] Confirm any required Datadog snapshot matches the selected scope
- [ ] Inspect Codex, Claude, and Cursor sessions for the same window
- [ ] Write `data/agent_sessions.json` with per-agent coverage and evidence
- [ ] Write `data/refresh_manifest.json` with a receipt for every source
- [ ] Write `data/personal_report.json` from the refreshed evidence
- [ ] Run `mise run report` to build HTML and the required one-page PDF
- [ ] Run `mise run verify-report`
- [ ] Ensure the local report server is serving the rebuilt report
- [ ] Inspect the served report at desktop and mobile sizes
- [ ] Render the PDF to PNG and inspect the complete single page
- [ ] Run `mise run check` if code or docs changed
- [ ] Verify the report uses current snapshot-backed evidence and narrative
```

## Default workflow

1. Start in this skill directory and work directly in the bundled runtime at
   `assets/activity-report-runtime`. Do not search the wider home directory for
   the runtime unless the user explicitly points you at another clone.
2. Read `README.md`, `report_config.py`, and `report_sources.py`. Read
   `references/personal-vs-org-reports.md` when the request asks for a personal
   report, when the user contrasts personal and org coverage, or when you need
   to export a single-page PDF. Read `references/visual-design-contract.md`
   whenever the user asks for a new UI, redesign, retheme, coat of paint, or a
   distinctly different look. Read `references/runtime-contract.md` when you
   need the exact date, snapshot, tracked source, or verification contract.
3. Decide whether this run is personal or org-wide before refreshing data.
4. Lock and announce the exact reporting window as described above. Keep the
   same `REPORT_DATE` and `REPORT_TIMEZONE` in the environment for every task.
5. If this is a redesign, complete the baseline audit and design brief before
   changing markup, CSS, or assets. The evidence refresh still happens in the
   same run unless the user explicitly asks for a visual-only prototype.
6. If `tracked_sources.json` is missing, run
   `mise run bootstrap-tracked-sources` before doing anything else. That task
   can seed the config from local snapshots or `dist/summary.json`.
7. Only if the bootstrap task says there were no usable local artifacts should
   you stop and tell the user to copy `tracked_sources.template.json` to
   `tracked_sources.json`, then fill in their private Slack channels and Notion
   pages.
8. Unless the user explicitly asked to use cache or not clear it, run
   `mise run clear-cache` before the refresh.
9. Treat cache as the generated report state:
   - clear `data/*.json`

From chadxz/prompts, MIT.

More from chadxz/prompts