Skip to content

codemap.md

Latest commit

 

History

History
42 lines (30 loc) · 1.42 KB

codemap.md

File metadata and controls

42 lines (30 loc) · 1.42 KB
Raw

packages/agent-events/

Speech pools and validation for agent feedback messages.

Responsibility

Provides categorized speech pools (thinking, success, error, permission) and validation logic for agent-facing messages. Ensures messages are safe (no code, URLs, paths, secrets) and appropriately sized (1-140 chars, single line).

Design

Category-Based Pools: Four speech categories with curated message pools:

  • thinking: "Thinking it through", "Let me check", etc.
  • success: "Done", "That worked", etc.
  • error: "Something failed", "Needs another look", etc.
  • permission: "Approval needed"

Validation Strategy: Regex-based validation rejecting:

  • Multi-line content (\r|\n)
  • Code-like patterns (backticks, keywords, braces)
  • URLs (https?://, www.)
  • File paths (slashes, drive letters)
  • Secrets (api_key, secret, password, token)

Random Selection: pickHookSpeech() uses bounded random index selection with fallback chains.

Flow

Agent Event → pickHookSpeech(category, randomFn) → validateHookSpeech(message) → Safe message

Integration Points

Consumers:

  • @open-pets/claude - Hook message generation
  • @open-pets/opencode - Plugin speech events

Exports:

  • hookSpeechPools - Readonly record of message arrays
  • pickHookSpeech() - Random selection with bounds checking
  • validateHookSpeech() - Security validation
  • HookSpeechCategory - Type union for categories