Custom bots and policies

In brief: Create named Meeting Bot personalities with a job role and versioned policy markdown so @mentions behave like the teammate you intended—QA, FDE, pre-sales, support, and more.

Meeting Bots start with seeded personalities (QABot, SalesBot). The extended model is to define custom bots with bespoke personalities and objectives, attach the right assistant(s) to each meeting, and brainstorm with agents that each bring a different lens—while TestChimp still supplies shared product / SDLC context (stories, scenarios, tests, releases, KBs, issues, observability, API schemas, and more).

Use cases go well beyond QA. Examples:

• FDE assistance — unblock implementation questions, point at plans, APIs, and coverage gaps
• PM critique — pressure-test scope, acceptance criteria, and edge cases from a product angle
• Pre-sales — answer with product truth without inventing capabilities you do not ship
• Customer support — ground troubleshooting in real issues, releases, and scenarios your team already tracks
• Domain / vertical bots — encode a specific mandate (compliance lens, performance risk, onboarding specialist) as a versioned policy

Create a bot

1. Open Meeting Bots → Bots.
2. Click Create, enter a name (unique, no spaces—e.g. FDEAssist, PMCritique).
3. Open the bot and set a Job role (short label shown in the UI, e.g. Senior FDE, Product Manager).
4. Edit the policy markdown and Save.

Joining: from the bot card, Join Meeting, pick a calendar event (preferred) or paste a Meet / Zoom / Teams / Webex URL.

In chat, org members invoke with @testchimp (default / general) or @YourBotName. Mentions are gated to org members (email preferred; exact display-name fallback when the meeting platform hides email).

What goes in the policy file

Each bot owns a policy file under:

plans/knowledge/policies/meetbots/<BotName>.policy.md

Draft shape (seeded when you create the bot):

---
bot-name: QABot
version: 1.0.0
---

### Policy

Describe how this personality should behave when mentioned in a meeting.

Piece	Guidance
Frontmatter bot-name	Match the bot’s display name (no spaces).
Frontmatter version	Bump when behaviour guidance changes so you can audit what the agent used.
### Policy body	Behavioural instructions: tone, what to prioritize, what to refuse, how to use TestChimp context, when to create issues / stories vs answer only.

The job role field on the bot record is complementary metadata (who this personality is). The policy body is the playbook the LLM follows on @mention.

Keep policies in Git with the rest of plans/knowledge/policies/ so teammates review personality changes like any other agent governance file. Same spirit as workflow policies—versioned guidance, not ad-hoc prompt paste.

Best practices

Be specific about the job. “Help in meetings” is weak. Prefer: Challenge ambiguous acceptance criteria; prefer citing existing scenarios; propose story drafts only when the team agrees; never invent APIs that contradict the plan.

One personality, one mandate. Split FDE assist and adversarial PM critique into two bots instead of one overloaded policy. Mentions stay predictable: @FDEAssist vs @PMCritique.

Tell the bot when to act vs answer. Spell out whether it should create issues, assign owners, or only recommend next steps until someone confirms.

Reuse product language. Point at how your team names stories, environments, and releases so answers align with QA Brain entities people already recognize.

Version on meaningful change. If you tighten red-team behaviour or change “always draft a story,” bump version so later audits know which guidance was live.

Prefer calendar-linked joins for @mentions. Calendar events improve participant email availability for the team guard. Ad-hoc pasted URLs can still transcribe; mentions may stay denied until emails are known—see Calendars.

Remember: one transcript. Multiple bots share the meeting transcription. Cost and storage scale with meeting time and LLM invokes, not with how many personalities you attached—see Billing.

Multi-bot meetings

Attach several personalities to the same call when you want structured brainstorming: one bot defends feasibility, another argues for the user, another tracks QA risk. Everyone still sees a single TestChimp Agent in the participant list; chat @BotName selects which policy answers.

After the call, use the shared transcript under Meetings for redaction, summarization, and ChimpHands follow-ups (Meetings).