Skip to content

READ.md

Latest commit

 

History

History
84 lines (69 loc) · 2.54 KB

READ.md

File metadata and controls

84 lines (69 loc) · 2.54 KB

Read from Agent Memory

When to Query

ALWAYS query at these task boundaries:

  • Starting a task - Scan for related solutions, patterns, intent
  • Planning work - Check architectural patterns, constraints, known issues
  • Writing tests - Look for testing patterns, common failures
  • Finishing a task - Review if your approach matches known intent
  • Writing commits - Verify your changes align with system intent
  • Documenting - Include discovered intent from memory
  • Context switching - Query new domain before starting

Query these domains before touching:

  • Authentication, authorization, sessions, cookies, tokens
  • API routes, endpoints, middleware, validation
  • Database models, migrations, queries
  • UI components, styling, theming (Gen3 design system)
  • Build config, deployment, CI/CD
  • Payment flows (Stripe, checkout)
  • Feature flags, A/B tests
  • Error handling, logging
  • React state management, hooks
  • Node modules, dependencies, package management

How to Query

# Search by keyword
jq -c 'select(.tags[] | test("KEYWORD"; "i"))' \
  $AGENT_SWARM_PATH/logs/*.jsonl | \
  jq -r '"[\(.timestamp | split("T")[0])] \(.repo): \(.lesson)"' | \
  tail -10

# Search by context (what was being done)
jq -c 'select(.context | test("YOUR_QUERY"; "i"))' \
  $AGENT_SWARM_PATH/logs/*.jsonl | \
  jq -r '"[\(.timestamp | split("T")[0])] \(.lesson)"' | \
  tail -10

# Search specific repo
jq -c 'select(.repo == "gptcoach-api")' \
  $AGENT_SWARM_PATH/logs/*.jsonl | \
  jq -r '"[\(.event_type)] \(.lesson)"' | \
  tail -10

Quick Queries

Before npm install:

jq -c 'select(.tags[] | test("npm"; "i"))' \
  $AGENT_SWARM_PATH/logs/*.jsonl | \
  jq -r '"\(.lesson) (\(.success_rate))"' | tail -5

Before touching API routes:

jq -c 'select((.tags[] | test("api"; "i")) and .type == "intent")' \
  $AGENT_SWARM_PATH/logs/*.jsonl | \
  jq -r '"\(.context)\n  → \(.lesson)\n"' | tail -20

Find architectural patterns:

jq -c 'select(.type == "pattern" or .type == "intent")' \
  $AGENT_SWARM_PATH/logs/*.jsonl | \
  jq -r '"[\(.repo)] \(.lesson)"' | tail -15

What You'll Get

Each result shows:

  • Date - When this was learned
  • Repo/System - Where it applies
  • Lesson - The actual knowledge
  • Success rate - How reliable (if logged)

Query Tips

  • Search before you make the same mistake twice
  • Use specific keywords: "build", "deploy", "auth", "api", "gen3"
  • Look for type == "intent" to understand WHY code exists
  • Check success_rate - high numbers mean proven solutions