@eddacraft/kindling-adapter-opencode

OpenCode session adapter for Kindling - capture tool calls, commands, and file changes from AI coding sessions.

Installation

npm install @eddacraft/kindling-adapter-opencode

Overview

Captures observations from OpenCode development sessions for local memory and continuity.

What is Captured

The OpenCode adapter automatically captures the following types of events from your development sessions:

Tool Calls

• Tool name and arguments
• Results or errors
• Duration (execution time)
• Timestamp

Command Execution

• Command text (e.g., git status, npm test)
• Exit code
• stdout and stderr output
• Working directory

File Changes

• File paths modified
• Diff content (additions/deletions)
• Change summary (lines added/deleted)

Errors

• Error message
• Stack trace preview
• Error source (runtime, validation, etc.)

Messages

• User messages (prompts, questions)
• Assistant messages (responses, explanations)
• Message length and model used

What is NOT Captured

Session lifecycle events (session_start, session_end) are skipped - they are used to manage capsules but not stored as observations.

Safety & Privacy

Automatic Redaction

The adapter automatically detects and redacts sensitive information:

• API keys, tokens, passwords: Patterns like api_key=, token:, password= are detected and values replaced with [REDACTED]
• AWS credentials: AWS secret access keys are masked
• Bearer/Basic auth: Authorization headers are sanitized
• Long secret-like strings: 32+ character alphanumeric strings with mixed letters/numbers are flagged as potential tokens

Excluded Files

The following file paths are automatically excluded from capture:

• node_modules/ directories
• .git/ directories
• .env files
• .pem, .key certificate files
• Files containing credentials or secrets in the path

Content Truncation

Large outputs are automatically truncated to 50,000 characters to prevent excessive storage usage. A truncation notice is appended when content is shortened.

Usage

Starting a Session

import { SessionManager } from '@eddacraft/kindling-adapter-opencode';
import { SqliteKindlingStore, initializeDatabase } from '@eddacraft/kindling-store-sqlite';

// Initialize store
const db = initializeDatabase(':memory:');
const store = new SqliteKindlingStore(db);

// Create session manager
const manager = new SessionManager(store);

// Start session
const context = manager.onSessionStart({
  sessionId: 'session-123',
  intent: 'Fix authentication bug',
  repoId: '/home/user/my-project',
});

Processing Events

// Process tool call event
manager.onEvent({
  type: 'tool_call',
  timestamp: Date.now(),
  sessionId: 'session-123',
  toolName: 'read_file',
  args: { path: 'src/auth.ts' },
  result: 'file contents...',
});

// Process command event
manager.onEvent({
  type: 'command',
  timestamp: Date.now(),
  sessionId: 'session-123',
  command: 'npm test',
  exitCode: 0,
  stdout: 'All tests passed',
});

Ending a Session

// End session with optional summary
manager.onSessionEnd('session-123', {
  reason: 'completed',
  summaryContent: 'Fixed JWT validation in auth middleware',
  summaryConfidence: 0.9,
});

Content Filtering

import { filterContent, truncateContent, maskSecrets } from '@eddacraft/kindling-adapter-opencode';

// Apply all safety filters
const filtered = filterContent(content, {
  maxLength: 10000,
  maskSecrets: true,
  showTruncationNotice: true,
});

// Just truncate
const truncated = truncateContent(longContent, { maxLength: 5000 });

// Just mask secrets
const masked = maskSecrets(contentWithSecrets);

Configuration

Currently, safety filters are applied by default and cannot be disabled. Future versions may add opt-in/opt-out configuration.

Data Storage

All captured observations are stored locally in SQLite via @eddacraft/kindling-store-sqlite. No data is sent to external services.

Observations are:

• Deterministically ordered by timestamp and sequence number
• Scoped to session, repository, agent, and user
• Queryable via full-text search and filters
• Redactable after capture if needed

Privacy Considerations

What you should know:

1. Local only: All data stays on your machine in a local SQLite database
2. Automatic sanitization: Sensitive data is detected and redacted automatically
3. Manual review: You can inspect and redact observations after capture
4. Export/forget: Observations can be exported or permanently deleted

What you should check:

• Review captured observations periodically for accidentally captured secrets
• Use /memory forget <id> to redact specific observations
• Consider excluding sensitive repositories from capture

License

Apache-2.0