Doordash — PR Scan Guide
Owner: Brylle Girang
Created: 2026-04-01
Cadence: Weekly (every Thursday)
Related: Doordash Workflow Plan · Packaging Guide
Purpose
This guide tells L&D how to run the weekly GitHub PR scan — the automated intake stage of the Doordash workflow. Engineers do not notify L&D when something ships. L&D finds it themselves by scanning merged PRs. This is the source of all Doordash deliveries.
What to scan
Repository: brainforge-ai/brainforge-platform
Scan window: All PRs merged in the last 7 days (Monday–Sunday of the preceding week).
Signal paths — filter for PRs that touch any of these:
| Path | Why it matters |
|---|---|
.cursor/skills/ | New or updated Cursor skills the team should use differently |
.cursor/rules/ | New or changed agent rules that affect how Cursor behaves |
knowledge/ | Vault taxonomy changes, new templates, new reference docs |
apps/platform/ | Forge-facing features, dashboard changes, UI updates team members interact with |
standards/ | Updated standards the team is expected to follow |
Ignore: PRs that only touch apps/platform/src/ internal implementation with no team-facing behavior change, purely cosmetic doc reformats, and dependency bumps with no workflow impact.
How to run the scan
Option A — Manual (use until automation is live)
- Open GitHub:
https://github.com/brainforge-ai/brainforge-platform/pulls?q=is%3Apr+is%3Amerged - Filter by date: set “Merged” to the past 7 days
- For each PR, open it and check:
- Which files changed (look at the “Files changed” tab)
- Does any changed file fall in the signal paths above?
- If yes → read the PR description and extract classification context (see below)
- Build a list of relevant PRs with tier assignment — this feeds directly into the packaging guide
Option B — Cursor background agent (scheduled, Thursday 9am)
See Cursor Automation Setup below. Once live, the agent runs the scan and posts the weekly L&D brief to #learning-and-development automatically. L&D reviews the brief and proceeds to packaging for any Tier 2 or Tier 3 items.
Extracting classification context from a PR
The existing PR description format (Summary, Changes, Impact, Related) contains everything L&D needs. No engineer follow-up required.
| PR field | What L&D extracts |
|---|---|
| Title | The one-line summary for the weekly digest or Slack post |
| Summary | What changed and why — the framing for packaging |
| Changes | Specific files modified — confirms which signal path was touched |
| Impact | Who is affected — sets the audience and targeting for delivery |
| Related | Linear ticket or issue ID — useful for service-line targeting |
| Files changed tab | Confirms the path; if .cursor/skills/ → high relevance regardless of description quality |
If a PR has no description: Flag it as a low-confidence classification. Either skip it for this week or check with the PR author for a one-sentence summary before packaging.
Classification rules
Classify each relevant PR into one of three tiers:
| Tier | Criteria | Examples |
|---|---|---|
| Tier 1 — Quick Tip | Single file change in .cursor/skills/ or .cursor/rules/; behavior change is one step different; no new workflow introduced | Renamed skill, updated skill prompt, new rule parameter |
| Tier 2 — Workflow Update | Multi-file change or change to an existing workflow; team members need to do something differently | Updated EP audit skill + rule, new Slack command added to a skill, vault taxonomy rename |
| Tier 3 — New System | Net-new capability the team has never had before; requires a full walkthrough to use correctly | New Forge feature, new agent workflow, new service-line tool added to the repo |
When in doubt, tier down. A Tier 1 misclassified as Tier 2 wastes packaging time. A Tier 2 misclassified as Tier 1 is correctable next week. A Tier 3 misclassified as Tier 2 is the only real risk — when uncertain between 2 and 3, ask: does someone need to see this demonstrated before they can use it?
Weekly scan output format
After scanning, produce a brief internal list (does not go to the team — this is L&D’s production note):
Week of [DATE]
TIER 1 (→ weekly digest)
- PR #[number]: [title] — [one line description of change]
TIER 2 (→ standalone Slack post, within 5 business days)
- PR #[number]: [title] — [one line description of change] | Audience: [all / SL name]
TIER 3 (→ Forge page + Zoom Clip, schedule with team)
- PR #[number]: [title] — [one line description of change] | Audience: [all / SL name]
SKIPPED (not team-facing)
- PR #[number]: [title] — reason skipped
This list is what gets packaged. See Packaging Guide for next steps.
Cursor Automation Setup
One-time configuration (Cursor web UI)
The Thursday automation runs the full Doordash workflow: scan → classify → format → post to Slack → log to changelog.
Setup steps:
- Open cursor.com → sign in → navigate to Background Agents
- Click New Schedule
- Configure:
- Name:
Doordash — Weekly L&D Brief - Schedule: Every Thursday at 9:00 AM (your timezone)
- Workspace:
brainforge-platform - Prompt: paste the master prompt below
- Name:
Master Thursday prompt (canonical version)
Run the weekly Doordash L&D production brief for the Brainforge team.
1. Read knowledge/people/learning-development/programs/doordash/doordash-pr-scan-guide.md
— follow it exactly to scan merged GitHub PRs from the past 7 days and classify each relevant update by tier.
2. Read knowledge/people/learning-development/programs/doordash/doordash-packaging-guide.md
— use the tier definitions, message format template, and per-tier packaging instructions to format the brief.
3. Post the formatted brief to #learning-and-development using the Slack MCP. Do not DM — post to the channel.
4. Append a new row to knowledge/people/learning-development/programs/doordash/doordash-changelog.md
logging today's date, the number of updates per tier, and the Slack post timestamp.
The prompt is intentionally short. All scan logic, tier definitions, and format instructions live in this guide and the packaging guide. When the process changes, update those docs — not the prompt.
Optional reminder automations
Tier 1 digest reminder (same Thursday, ~1 hour after master):
Read knowledge/people/learning-development/programs/doordash/doordash-changelog.md.
Check whether this week's Tier 1 items have been logged as packaged into a digest.
If not, post a reminder to #learning-and-development to publish the "This week in the platform" digest before EOD today.
Tier 2 SLA follow-up (Tuesday after each Thursday brief):
Read knowledge/people/learning-development/programs/doordash/doordash-changelog.md and
knowledge/people/learning-development/programs/doordash/doordash-sla.md.
Identify any Tier 2 items from the last Thursday brief that are now past their 5-business-day packaging SLA and have not been logged as delivered.
Post a reminder to #learning-and-development listing the overdue items and their deadlines.
Troubleshooting
No PRs this week: Post the digest anyway with a “quiet week” note. Skipping the digest breaks the cadence.
PR with no description: Skip for now; flag to author for a one-sentence summary. Do not invent context.
Multiple PRs touching the same skill: Combine into one update. One skill = one delivery, even if two PRs touched it.
PR merged outside the 7-day window: If it was missed in a previous scan, include it in the current week’s output with a note (“missed last week”).