Cross-Cutting & Pod Support — Mandate Plans
Mandates PS-040 through PS-043 (principles and rhythms) and POD-001 through POD-006 (small builds and training).
PS-040: Human-in-the-Loop on All Writes
What this means
This isn't a task; it's a design principle that governs everything you build. Any automation that touches pay, leave status, or employee records must follow the propose-then-approve pattern:
- Agent processes data and proposes a change
- Human reviews the proposal
- Human approves or rejects
- Only approved changes are written to the system of record
This is non-negotiable. The existing agents all follow this pattern (the recon agent flags findings for Syra to review; the leave agent proposes Paylocity re-entries for Grecia to approve). Everything you build must do the same.
Concrete tasks
- Review how the existing agents implement HITL (Human-in-the-Loop): the Exception Queue in the recon agent, the notification flow in the leave agent.
- Document the HITL pattern as a reusable design template for new agents.
- For every automation you build in September, verify: can the agent write to a system of record without human approval? If yes, fix it.
PS-041: Join People Team Calls
What this means
Show up to the team's existing meetings. Learn how they communicate, what they care about, what frustrates them. This is relationship building, not automatable.
Concrete tasks
- Get on the team's meeting calendar in your first week. Ask Shlanda or Syra what recurring meetings you should attend.
- Understand the standing rhythms: weekly team sync, Wednesday working block, any ad-hoc calls.
- Listen more than talk in the first two weeks. Note pain points, repeated complaints, and workarounds.
People to meet
| Person | Questions to ask |
|---|---|
| Shlanda | What meetings should I be on? What's the team's communication style (Slack-first, email, meetings)? What should I know about team dynamics? |
PS-042: Wednesday Working Block
What this means
The standing Wednesday 9-10:30am CT session is the primary training and transfer mechanism. This is where Shlanda's team learns to run the agents, use the department brain, and extend the Slack bot.
Concrete tasks
- Prepare a topic for each Wednesday (not ad hoc; have a plan).
- August Wednesdays: demo the existing agents, show the team the department brain, walk through a live exception review.
- September Wednesdays: pair-program on new builds, show the team how to read rules, modify config.
- October Wednesdays: the team runs the session; you observe and answer questions.
- Track skill transfer: who can do what by the end of each month?
PS-043: Scope and Build Team-Wide Automations
What this means
Beyond your core People Systems seat work, you scope and build additional automations that benefit the broader team. These are prioritized with Shlanda inside the monthly cadence.
Concrete tasks
- Maintain a backlog of automation requests from the team (Shlanda, Syra, Nickole, Grecia).
- Prioritize by impact and effort each month with Shlanda.
- Build the highest-value items within the $6K/month pod support scope.
POD-001: DocuSign Audit (~5,700 files)
What this involves
MR has ~5,700 DocuSign documents that need to be sorted and organized. They're in a Drive folder but not categorized by document type, signer, date, or status.
Concrete tasks
- Assess the scope: how are the 5,700 files currently organized (or not)? What categories do they need to be sorted into?
- Build a classification agent: read each document's metadata (signer, date, document type from title/content) and sort into folders or tag in a spreadsheet.
- Set up ongoing routing: new DocuSign completions auto-sort into the right category.
Dependencies
- DocuSign API access (credentials are in
secrets.env: integration key, user ID, secret key, RSA private key) - DocuSign inbox Drive folder (
0AG5VbV08w_L3Uk9PVAfromsecrets.env) - Classification taxonomy from Shlanda or the team (what categories exist?)
People to meet
| Person | Questions to ask |
|---|---|
| Shlanda | What categories should DocuSign documents be sorted into? What's the current mess look like? Who needs to find these documents and for what? |
| Kyle | DocuSign API is set up in secrets.env. Has anyone tested it? Is the JWT auth flow working? |
POD-002: DocuSign Ongoing Sorting
What this involves
After the initial audit (POD-001), set up automated sorting so new documents route correctly without manual intervention.
Concrete tasks
- Build a webhook listener or scheduled poller using DocuSign Connect (webhooks are well-supported; see
systems/docusign.md). - When a new document is completed: classify it, file it in the right Drive folder, log it in a tracking spreadsheet.
Dependencies
Same as POD-001, plus DocuSign Connect webhook configuration.
POD-003: Creative Brief Bot
What this involves
A bot that helps Denise's (marketing creative) team generate creative briefs. It's already referenced as being "in front of" the team. May serve as a template for similar bots in People ops.
Concrete tasks
- Understand the current state: is this already built? By whom? Where does it live?
- Assess if the People team needs a similar pattern: brief generation for training materials, onboarding guides, policy summaries.
People to meet
| Person | Questions to ask |
|---|---|
| Kyle or Mikyo | Where does the creative brief bot live? Is it an Apps Script, Claude Code artifact, or something else? Can the pattern be reused? |
POD-004: HR Inbox Triage
What this involves
The HR inbox gets ~13,000 emails per year routed to 5 people. Manual sorting; no smart routing. A triage bot was targeted but dropped from the SOW (Brad deferred June 3; Shlanda prefers build-in-house).
It's listed in the small-build lane but at low priority. If addressed, a quick-win Google Apps Script bot hitting the Anthropic API could handle ~30% of inbox volume (~2 days of work).
Status
OUT of SOW. Only build this if explicitly asked by Shlanda. Don't prioritize it over the core mandates.
If it comes back into scope
- Classify the inbox: what types of emails come in? What percentage are routine policy questions vs. ER cases vs. benefits questions?
- Build a Tier-1 deflection bot: route common policy questions to an auto-response; flag ER/EEOC cases for immediate human attention.
- Critical safety gate: ER and EEOC cases must NEVER get an automated response. The routing must be conservative (false negative = human reviews a routine question; false positive = ER case gets auto-responded, which is unacceptable).
POD-005: Weekly Training Sessions
What this involves
The Wednesday 9-10:30am CT working block (same as PS-042). You and Kyle facilitate; the team learns to own the platform.
Concrete tasks
See PS-042 above. The key addition from the pod support perspective:
- Track skill transfer quantitatively: by October, each team member should be able to perform specific tasks independently (Nickole extends the Slack bot, Syra reviews the Exception Queue, Shlanda understands the architecture).
POD-006: Mentor-Layer Backing
What this involves
Escalation support and office-hours pairing for Shlanda, Nickole, and Syra. When they get stuck on the agents or the department brain, you (and Kyle) are available to unblock them.
Concrete tasks
- Set up a "Daniel is available" signal: a Slack channel, office hours, or just "DM me anytime."
- Pair-program with Nickole at least once per week (she's the AI-Ops candidate; building her skills is strategic).
- Help Shlanda design the pod/cohort structure for how the broader team will learn AI tools.
- Track escalation patterns: what do people get stuck on most? That tells you what needs better documentation or simpler interfaces.
People to meet
| Person | Questions to ask |
|---|---|
| Nickole | What's your current comfort level with GitHub, Claude Code, the department brain? What do you want to learn? What's your biggest blocker? |
| Syra | Same: what would help you feel more confident with the tools? Where do you get stuck? |
| Shlanda | What does the mentor layer look like after October? Who on your team is closest to self-sufficient? |