# AGENTS.md — Build guide for Chatbase

## Project scope
Ingest a bounded set of permitted documents, review extraction and expose a support widget that retrieves relevant passages. Add one allowlisted read-only account lookup and a human handoff when evidence is insufficient.

Catalogue verdict: kinda. A useful web version is a serious multi-day build, not magic: crawl or upload sources, retrieve tenant-scoped passages, stream cited answers, run one guarded business action, and hand uncertain requests to a person. Rebuilding Chatbase itself is a different project. Its paid surface combines source management and retraining, procedures and custom UI actions, a polished widget, helpdesk, identity, analytics, voice, email, social and CRM channels, plus the reliability and security work that keeps an agent safe in front of customers.
Use the implementation prompt below to define the deliverable. Complete each phase's acceptance checks before extending the scope.

## Working agreement
- Inspect the repository and its existing instructions before choosing paths, dependencies or commands. Keep one coherent stack and explain changes to the proposed architecture.
- Plan a vertical slice that accepts a real input and produces the useful output described below. Persist only the state the prompt calls for; respect memory-only and upstream-managed workflows. Use fixtures only when they are clearly labelled.
- After scaffolding, document the actual install, development, check and build commands in README and keep them synchronized with the package or project manifest. Do not report commands as successful unless they ran.
- Work in small steps. At handoff, list implemented flows, checks actually performed, remaining blockers, and any credentials or provider setup the owner must supply.
- Do not publish, spend money, contact customers, delete source data or run irreversible migrations without the project owner's authorization.

## Prerequisites
- A supported Node release, PostgreSQL, HTTPS for a shared deployment, and a backup destination. Begin with one workspace and explicit owner/member permissions.
- Implementation components: Next.js App Router and TypeScript for server-rendered pages and validated mutations. PostgreSQL with Drizzle migrations; Better Auth sessions for a small private workspace.
- Scope boundary: Broad autonomous actions, enterprise connectors and guaranteed answer accuracy are outside scope.

## Stack and architecture
- Next.js App Router and TypeScript for server-rendered pages and validated mutations.
- PostgreSQL with Drizzle migrations; Better Auth sessions for a small private workspace.
- Domain model: approved knowledge sources, source revisions, chunks, conversations, citations and human handoff tickets

## Security and data integrity
- Authorize every record read and mutation on the server using its workspace membership; validate payloads, protect mutations against CSRF, and escape user-authored HTML.
- Correctness boundary: Knowledge text cannot change system/tool policy; every source citation belongs to the active knowledge revision and private documents stay tenant-scoped.
- Commit related database changes in one transaction. Use version checks for competing edits and an outbox for external notifications; retry delivery independently of saving the record.
- Export versioned JSON plus attachments and a readable CSV summary. Restore into a separate database and compare record IDs and attachment checksums before switching.

## Agent implementation rules
- Project rule — data model: approved knowledge sources, source revisions, chunks, conversations, citations and human handoff tickets
- Project rule — preserve this invariant: Knowledge text cannot change system/tool policy; every source citation belongs to the active knowledge revision and private documents stay tenant-scoped.
- Project rule — acceptance evidence: A document containing prompt-injection text cannot trigger a tool; an unsupported question creates a handoff with the selected transcript rather than an invented answer.

## Optional agent skills and references
- Optional external skill: [vercel-react-best-practices](https://github.com/vercel-labs/agent-skills/blob/main/skills/react-best-practices/SKILL.md) — Improve React and Next.js data fetching, rendering, bundle size and server performance. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.
- Optional external skill: [supabase-postgres-best-practices](https://github.com/supabase/agent-skills/blob/main/skills/supabase-postgres-best-practices/SKILL.md) — Review PostgreSQL schemas, queries, indexes, pooling, concurrency and row-level security. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.
- Optional external skill: [web-design-guidelines](https://github.com/vercel-labs/agent-skills/blob/main/skills/web-design-guidelines/SKILL.md) — Review web interfaces for accessibility, keyboard focus, forms, navigation and interaction quality. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.

Read the linked SKILL.md and its dependencies before adding a skill. Select only the skills matching this project's runtime and task; their documentation does not supply API access, credentials or approval to perform external actions. Pin the reviewed revision where the tool supports it. Follow the chosen agent's documented project-level installation mechanism.

## Distribution ideas
These are optional planning notes. Obtain the owner's approval before publishing or contacting anyone.
- Demonstrate the actual Chatbase-inspired workflow with owned or clearly labeled sample data: Ingest a bounded set of permitted documents, review extraction and expose a support widget that retrieves relevant passages. Add one allowlisted read-only account lookup and a human handoff when evidence is insufficient.
- Publish a reproducible walkthrough with this observable result: A document containing prompt-injection text cannot trigger a tool; an unsupported question creates a handoff with the selected transcript rather than an invented answer.
- Explain who can operate this scoped tool, its setup and ongoing costs, and these remaining product gaps: Broad autonomous actions, enterprise connectors and guaranteed answer accuracy are outside scope. Avoid guaranteed savings, performance scores or implied endorsement.

## Engineering roadmap
1. Phase 1 — Scope and fixtures. Implement this bounded workflow: Ingest a bounded set of permitted documents, review extraction and expose a support widget that retrieves relevant passages. Add one allowlisted read-only account lookup and a human handoff when evidence is insufficient. Record prerequisites, select representative user-owned fixtures and document the unsupported features: Broad autonomous actions, enterprise connectors and guaranteed answer accuracy are outside scope.
2. Phase 2 — Durable model. Model approved knowledge sources, source revisions, chunks, conversations, citations and human handoff tickets Add migrations or a versioned document format, explicit validation, stable IDs and a visible import-error report. Preserve this rule: Knowledge text cannot change system/tool policy; every source citation belongs to the active knowledge revision and private documents stay tenant-scoped.
3. Phase 3 — Complete the first useful path. Implement the workflow's input, review and output interface, with clear controls and explicit empty/error states. Commit related database changes in one transaction. Use version checks for competing edits and an outbox for external notifications; retry delivery independently of saving the record.
4. Phase 4 — Permissions and integration failure. Authorize every record read and mutation on the server using its workspace membership; validate payloads, protect mutations against CSRF, and escape user-authored HTML. Request integration credentials and permissions only for the enabled feature; show a disconnected state instead of mock results.
5. Phase 5 — Portable handoff. Export versioned JSON plus attachments and a readable CSV summary. Restore into a separate database and compare record IDs and attachment checksums before switching. Include setup, operating limits, fixture walkthrough and shutdown/restart instructions in the README.
6. Phase 6 — Acceptance scenarios. A document containing prompt-injection text cannot trigger a tool; an unsupported question creates a handoff with the selected transcript rather than an invented answer. Repeat the workflow after restart and with a denied permission or unavailable dependency; show recoverable failure rather than a success placeholder.

## Paid-product capabilities outside this build
- the no-code agent lifecycle: many managed source connectors, background retraining, source suggestions, and production ingestion diagnostics
- the channel network: website chat, email, voice and telephony, Slack, WhatsApp, Messenger, Instagram, Shopify, WordPress, Zapier, Zendesk, Salesforce, and other integrations
- the support operation around the agent: a team helpdesk, tickets, assignment, contact identity, authenticated personalization, escalation, and access controls
- the action and procedure builder: server, client, button, and custom widget actions with testing, permissions, and integration-specific behavior
- the production layer: topic and sentiment analytics, review workflows, abuse controls, delivery retries, observability, data governance, compliance, and support

## Implementation prompt
WORKING SLICE
Ingest a bounded set of permitted documents, review extraction and expose a support widget that retrieves relevant passages. Add one allowlisted read-only account lookup and a human handoff when evidence is insufficient.

Build this scoped Chatbase-inspired workflow with a documented data model and visible failure states.

Architecture
- Next.js App Router and TypeScript for server-rendered pages and validated mutations.
- PostgreSQL with Drizzle migrations; Better Auth sessions for a small private workspace.

Prerequisites and limits
A supported Node release, PostgreSQL, HTTPS for a shared deployment, and a backup destination. Begin with one workspace and explicit owner/member permissions.
Outside this release: Broad autonomous actions, enterprise connectors and guaranteed answer accuracy are outside scope.

Data model and correctness
approved knowledge sources, source revisions, chunks, conversations, citations and human handoff tickets
Invariant: Knowledge text cannot change system/tool policy; every source citation belongs to the active knowledge revision and private documents stay tenant-scoped.
Commit related database changes in one transaction. Use version checks for competing edits and an outbox for external notifications; retry delivery independently of saving the record.

Security and privacy
Authorize every record read and mutation on the server using its workspace membership; validate payloads, protect mutations against CSRF, and escape user-authored HTML.

Recovery and export
Export versioned JSON plus attachments and a readable CSV summary. Restore into a separate database and compare record IDs and attachment checksums before switching.

Implementation order
1. Phase 1 — Scope and fixtures. Implement this bounded workflow: Ingest a bounded set of permitted documents, review extraction and expose a support widget that retrieves relevant passages. Add one allowlisted read-only account lookup and a human handoff when evidence is insufficient. Record prerequisites, select representative user-owned fixtures and document the unsupported features: Broad autonomous actions, enterprise connectors and guaranteed answer accuracy are outside scope.
2. Phase 2 — Durable model. Model approved knowledge sources, source revisions, chunks, conversations, citations and human handoff tickets Add migrations or a versioned document format, explicit validation, stable IDs and a visible import-error report. Preserve this rule: Knowledge text cannot change system/tool policy; every source citation belongs to the active knowledge revision and private documents stay tenant-scoped.
3. Phase 3 — Complete the first useful path. Implement the workflow's input, review and output interface, with clear controls and explicit empty/error states. Commit related database changes in one transaction. Use version checks for competing edits and an outbox for external notifications; retry delivery independently of saving the record.
4. Phase 4 — Permissions and integration failure. Authorize every record read and mutation on the server using its workspace membership; validate payloads, protect mutations against CSRF, and escape user-authored HTML. Request integration credentials and permissions only for the enabled feature; show a disconnected state instead of mock results.
5. Phase 5 — Portable handoff. Export versioned JSON plus attachments and a readable CSV summary. Restore into a separate database and compare record IDs and attachment checksums before switching. Include setup, operating limits, fixture walkthrough and shutdown/restart instructions in the README.
6. Phase 6 — Acceptance scenarios. A document containing prompt-injection text cannot trigger a tool; an unsupported question creates a handoff with the selected transcript rather than an invented answer. Repeat the workflow after restart and with a denied permission or unavailable dependency; show recoverable failure rather than a success placeholder.

Acceptance
A document containing prompt-injection text cannot trigger a tool; an unsupported question creates a handoff with the selected transcript rather than an invented answer.
Use real source data or clearly labeled fixtures. Explain unsupported input and provider failures; do not fabricate analytics, delivery receipts, accuracy claims or security guarantees.

Optional agent guidance
Optional external skill: [vercel-react-best-practices](https://github.com/vercel-labs/agent-skills/blob/main/skills/react-best-practices/SKILL.md) — Improve React and Next.js data fetching, rendering, bundle size and server performance. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.
Optional external skill: [supabase-postgres-best-practices](https://github.com/supabase/agent-skills/blob/main/skills/supabase-postgres-best-practices/SKILL.md) — Review PostgreSQL schemas, queries, indexes, pooling, concurrency and row-level security. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.
Optional external skill: [web-design-guidelines](https://github.com/vercel-labs/agent-skills/blob/main/skills/web-design-guidelines/SKILL.md) — Review web interfaces for accessibility, keyboard focus, forms, navigation and interaction quality. Review its instructions and compatibility before use; it does not grant deployment, data-access or publication permission.
Project rule — data model: approved knowledge sources, source revisions, chunks, conversations, citations and human handoff tickets
Project rule — preserve this invariant: Knowledge text cannot change system/tool policy; every source citation belongs to the active knowledge revision and private documents stay tenant-scoped.
Project rule — acceptance evidence: A document containing prompt-injection text cannot trigger a tool; an unsupported question creates a handoff with the selected transcript rather than an invented answer.

## Completion evidence
Demonstrate the prompt's acceptance scenarios against the scoped workflow. Include setup from a clean checkout and failure recovery. Check persistence across restart and export/restore only for the state the prompt says to store; for memory-only tools, confirm that temporary content is discarded as specified. Record actual results and remaining limitations. A detailed plan alone does not establish a working replacement.