Tech Docs KB Search

Search technical and architectural documentation using AWS Bedrock Knowledge Base. Provides semantic and hybrid search over the lightspeed-hospitality/tech-docs corpus without needing a local clone.

Trigger

Use this skill when the user asks to:

• Find technical documentation, ADRs, RFCs, runbooks
• Understand service architecture or data flows
• Look up development guidelines or deployment procedures
• Run /tech-docs

Prerequisites & Authentication

• AWS CLI v2 (brew install awscli)
• granted CLI (brew install common-fate/granted/granted)
• Access to the lsk/dev/admin/eu-central-1 role

For prerequisite checks and SSO login flow, see references/auth.md.

IMPORTANT: Do NOT proceed with KB queries until credentials are confirmed working.

Configuration

IMPORTANT: All aws commands must include --profile "lsk/dev/admin/eu-central-1" --region eu-central-1.

Indexed Content

Not indexed: /blog content is intentionally excluded.

Search the Knowledge Base

Reformulate the Query

Do NOT pass the user's question verbatim. Extract key nouns and technical terms for HYBRID search:

Strategy: Include the topic, specific service/tool names, and 2-3 terms you'd expect in the answer document.

Run the Search

Always default to HYBRID search. It outperforms SEMANTIC-only in both precision and token efficiency.

aws bedrock-agent-runtime retrieve \
  --knowledge-base-id S3WJX83IGW \
  --retrieval-query '{"text": "QUERY"}' \
  --retrieval-configuration '{"vectorSearchConfiguration": {"numberOfResults": 5, "overrideSearchType": "HYBRID"}}' \
  --profile "lsk/dev/admin/eu-central-1" \
  --region eu-central-1 \
  --output json \
  --no-cli-pager

For metadata filters and advanced search patterns, see references/advanced-search.md.

Handle Errors

• ExpiredTokenException / Invalid credentials: Re-run auth (see references/auth.md), then retry
• AccessDeniedException: User may not have the required role
• ThrottlingException: Wait 3 seconds and retry once

Parse and Present Results

Evaluate Result Quality

Process Results

1. Read chunks from retrievalResults[].content.text
2. Extract source paths from retrievalResults[].location.s3Location.uri — strip bucket prefix to get repo-relative path
3. For high-relevance results (score >= 0.4), fetch the full document from S3 — see references/advanced-search.md
4. Synthesize an answer — do NOT dump raw JSON or chunk text
5. Cite sources with repo-relative file paths

Presentation Format

## [Topic]

[Synthesized answer from retrieved chunks and full documents]

**Sources:**
- `docs/path/to/file.md` — [what this doc covers]

Checklist

• Verified AWS CLI v2 and active credentials before searching
• Reformulated the query (extracted key terms, not verbatim)
• Used HYBRID search as default
• Actually ran the aws command using the Bash tool
• Evaluated result scores and retried if top score < 0.4
• Fetched full documents from S3 for high-relevance results
• Synthesized an answer (not raw dump)
• Cited source document paths