Enrichment Agent: Architecture & Trust

Nodes (main)

Data Sources

Trigger Modes

Dedup Levels

Active Workflows

Workflow 1: Enrichment Agent

Linear Ticket Enrichment Agent v3.1 ActiveG3TTZLOOX56FOv42

Linear Trigger

→

Validate & Route

→

AI Agent (Claude 4.6)

→

Post Comment

11 tool nodes connected to AI Agent via ai_tool | Timezone: America/New_York | Model: claude-sonnet-4-6, temp 0.2

Trigger Modes

Mode	Trigger Condition	Enrichment Scope	Dedup Window	Report Tag
create	New D2C ticket created (human actor only)	Full protocol: all 11 sources, claim verification, follow-the-leads	1 hour	(none)
status_in_progress	Ticket moves to any started state + updatedFrom.stateId present	Full protocol (same as create). Developer needs fresh context	7 calendar days	(re-enrichment)
pr_linked	Any update on ticket with GitHub PR in attachments	Lightweight: PR details + brief Slack (2 queries) + related tickets only. Skips BQ, Sentry, Vercel, Drive, Notion, Figma	24 hours	(lightweight)

11 Connected Data Sources

🐙

GitHub

PRs, commits, reviews

💬

Slack

Search, threads, users

✎

Linear

Issues, comments, search

🐛

Sentry

Errors, stacktraces

▲

Vercel

Deploys, build logs

📑

Notion

Wiki, docs, pages

📂

Google Drive

Specs, PRDs, docs

📊

BigQuery

Metrics, claim verify

🎨

Figma

Nodes, comments

📈

Langfuse

LLM traces, prompts

⚙

Statsig

Gates, experiments

Source routing: Base layer (GitHub, Slack, Linear, Drive, Notion) runs on EVERY ticket. Specialized layer (BQ, Sentry, Vercel, Figma, Langfuse, Statsig) activates by ticket type + content triggers. Lightweight mode skips specialized layer entirely.

3-Level Deduplication

Level 1
In-Memory Lock
globalThis lock per issueId. Blocks duplicate webhooks within 120 seconds. First line of defense, catches ~90% of duplicates instantly.
Level 2
Comment Age Check
Queries Linear comments in Validate node. Checks report age against trigger-specific window (1h / 7d / 24h). Prevents unnecessary AI Agent runs.
Level 3
Database Dedup
Post Comment node queries Linear before posting. Blocks if report posted <5 min ago. Catches race conditions from parallel n8n workers.

Workflow 2: Stale Ticket Detector

Stale Ticket Detector v1.0 ActiveJtmMZikcr83HiEi7

Daily Schedule (10 AM ET, weekdays)

→

Detect Stale Tickets

→

Post Health Check

Scans D2C "In Progress" tickets | Flags if no activity for 7+ days | Health Check dedup: 24 hours

Parameter	Value
Target	D2C team tickets in started state type
Staleness threshold	7 calendar days since last updatedAt
Health Check dedup	Skip if ## Health Check comment posted < 24 hours ago
Comment includes	Days stale, assignee, linked PR count, actionable suggestions
Schedule	Weekdays at 10:00 AM ET (cron: 0 10 * * 1-5)

Enrichment Report Structure

## Enrichment Report

Type: bug/feature/analytics/infra | Confidence: High/Medium/Low | Run: timestamp ET (re-enrichment)

### TL;DR

2-3 sentences. Lead with most important finding. Actionable direction.

### Code

PRs with size metrics, review status, author. One line each.

### Errors

Sentry issues (top 2 by event count). Bug type only.

### Discussions

Verified Slack messages. Grouped by topic if 4+.

### Docs

Google Drive + Notion docs. Content-verified before inclusion.

### Design

Figma nodes with variant summary. Feature type only.

### Claim Verification

Every claim: Confirmed / Contradicted / Unverified + concrete evidence.

### Related Tickets

Duplicates, prior art, parallel work, blockers.

### Open Questions

Max 1-2 genuinely unresolved items.

Validate Node Decision Tree

1. Extract issueId from webhook payload

2. Check in-memory lock (120s per issueId) → skip if locked

3. Determine action:

create → filter bot actors → triggerReason = "create"

update + state started → filter bot actors → triggerReason = "status_in_progress"

update + other → triggerReason = "_check_pr_linked"

other action → skip

4. Fetch full issue via Linear GraphQL

5. Filter: team must be D2C, no "skip-enrichment" label

6. Find age of most recent Enrichment Report comment

7. Apply trigger-specific dedup:

create: report < 1h → skip

status_in_progress: report < 7 days → skip

_check_pr_linked: no GitHub PR → skip | report < 24h → skip

8. Pass to AI Agent with issueData + triggerReason + ET timestamp

Production Validation

Test	Ticket	Result	Duration
create Full enrichment	D2C-4849	Pass 3 PRs, Statsig gate, related tickets	~85s
re-enrichment	D2C-4849	Pass "(re-enrichment)" tag, fresh data	~85s
pr_linked Lightweight	D2C-4766	Pass PR #1138, 3 Slack, 3 related	~36s
Dedup: fresh report	D2C-4849	Pass In Progress <7d skipped	<1s
Race condition fix	Post Comment	Fixed 5-min DB dedup	n/a
Figma integration	D2C-4828	Pass URL parse, node, comments	~56s

Required n8n Configuration

Variables (16)

Variable	Source	Used By
LINEAR_API_KEY	Linear Settings > API	Validate, Linear API, Post Comment
LINEAR_OAUTH_CLIENT_ID	Linear OAuth App	Post Comment (bot identity)
LINEAR_OAUTH_CLIENT_SECRET	Linear OAuth App	Post Comment (bot identity)
GITHUB_TOKEN	GitHub PAT	GitHub API
SLACK_BOT_TOKEN	Slack App > OAuth	Slack API
SENTRY_AUTH_TOKEN	Sentry > Auth Tokens	Sentry API
VERCEL_TOKEN	Vercel > Tokens	Vercel API
NOTION_API_KEY	Notion Integration	Notion API
GOOGLE_SA_CLIENT_EMAIL	GCP Service Account	Drive, BigQuery
GOOGLE_SA_PRIVATE_KEY_1	GCP SA key (part 1)	Drive, BigQuery
GOOGLE_SA_PRIVATE_KEY_2	GCP SA key (part 2)	Drive, BigQuery
GOOGLE_PROJECT_ID	GCP Project	BigQuery
FIGMA_API_TOKEN	Figma PAT	Figma API
LANGFUSE_PUBLIC_KEY	Langfuse Settings	Langfuse API
LANGFUSE_SECRET_KEY	Langfuse Settings	Langfuse API
STATSIG_API_KEY	Statsig Console	Statsig API

Credentials (2)

Credential	Type	Used By
Anthropic account	anthropicApi	Claude Sonnet 4.6
Linear account	linearApi	Webhook trigger