- Add deleteWorkspaceByChannelId() to DisclawDatabase
- Add src/lifecycle/categories.ts: ensureCategories() + moveToCategory()
- Add src/lifecycle/reconcile.ts: reconcile() assigns channels to active/inactive/archive categories on startup
- Add src/lifecycle/channel-delete-handler.ts: handles channelDelete events for management, category, and agent channels
- Wire lifecycle into bot.ts: state object, ensureCategories, reconcile, channelDelete listener
- Extend handleNewAgent() with optional activeCategoryId to place new channels in the active category
- Add 13 unit tests covering all lifecycle scenarios
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two root causes for the failing timeout test:
1. Test config used `workspace_root` instead of `workspaces_root`, causing
Zod validation to fail silently, so resolveTimeoutMs() fell back to the
default 120s — the mock's 200ms close event always beat it.
2. Schema enforced `.int()` on claude_timeout_seconds, rejecting the test's
0.05s fractional value. Removed the integer constraint since sub-second
timeouts are valid (and useful for testing).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add RunResult type in src/agent/types.ts (success|timeout|cli-error|parse-error).
Runner resolves instead of rejects for all error cases.
Router handles all four status variants explicitly.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add ChannelQueue in src/runtime/channel-queue.ts.
Router enqueues each message so same-channel requests run serially.
Different channels remain parallel.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The flush() helper now enforces the Discord hard limit by hard-splitting
any chunk that exceeds 2000 chars (e.g. when "\n```" was appended after
the soft-limit check). Also reserves 4 chars of overhead in the
soft-limit check when inside a fenced code block.
Adds a regression test that verifies no chunk exceeds 2000 chars even
when the fence closure is appended at the limit boundary.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Resolve conflict in src/router.ts: keep both splitForDiscord (DIS-105)
and sanitizeForDiscord (DIS-102). Apply splitForDiscord to the already
sanitized response text.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The root cause was twofold:
1. Raw (unsanitized) agent responses were stored in the conversation
history DB. On subsequent calls, these paths were fed back to the
agent as context, causing it to reproduce full paths regardless of
Discord-side sanitization.
2. repoRoot (workspace_path) was not passed to sanitizeForDiscord,
so workspace-specific paths were only partially masked via the
home directory fallback.
Fix: sanitize the full response once before both sending to Discord
and storing in the DB. Pass workspace_path as repoRoot for precise
path replacement. Add Windows path tests.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add sanitizeForDiscord() in src/runtime/sanitize.ts.
Router applies sanitizer to all channel.send() calls.
Prevents path/token leaks into Discord channels.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Move all CREATE TABLE/INDEX DDL from database.ts to schema.sql.
database.ts loads and executes schema.sql via db.exec().
Add integration test verifying idempotent schema init.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add DisclawConfigSchema and AgentYamlSchema with Zod validation.
loader.ts and identity.ts validate external data at parse time.
Clear error messages with field name and expected type.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add splitForDiscord() in src/discord/split.ts.
Fences are properly closed and reopened across chunk boundaries.
Router uses splitForDiscord instead of internal splitMessage().
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace all console.* calls in src/ with pino child-loggers.
Add rootLogger + childLogger(component) in src/runtime/logger.ts.
pino-pretty used as dev transport only.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add Semaphore class in src/runtime/concurrency.ts.
Runner acquires/releases slot around claude spawn.
Default cap: 4, configurable via disclaw.yaml max_concurrent_agents.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
--bare disables OAuth/keychain auth and requires ANTHROPIC_API_KEY,
breaking Claude Pro/Max subscriptions. CLAUDE.md isolation is now
achieved structurally: workspaces live under ~/.disclaw/workspaces/
(outside the repo) so walk-up never reaches DisClaw's CLAUDE.md.
Update integration test to assert --bare is absent and verify correct
args (-p, --output-format json). Fix sanitize-env tests to reflect
CI=true removal (was breaking OAuth headless-mode detection).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Implements acquireLock() in src/runtime/lockfile.ts that writes a PID-stamped
lock file under workspaces_root, detects live vs stale locks via process.kill(pid, 0),
and returns a release function. src/index.ts acquires the lock before DB init and
releases it in the graceful-shutdown handler. Four unit tests cover acquire, duplicate-
acquire rejection, stale-lock recovery, and release.
Prevents CLAUDE.md walk-up through parent directories by passing --bare
to every Claude CLI invocation, and explicitly injects the workspace
CLAUDE.md via --append-system-prompt-file so agent identity is isolated.
Adds fake-claude fixture and integration test to verify flags are present.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Introduces src/runtime/env.ts with a pure sanitizedEnv() function that
strips DISCORD_BOT_TOKEN, DISCORD_CLIENT_SECRET, DISCORD_PUBLIC_KEY, and
all keys matching /^(DISCLAW_SECRET_|SECRET_|TOKEN_|API_KEY_?)/i before
passing the environment to spawned Claude Code child processes. Adds
optional env_blocklist field to DisclawConfig. Runner now passes
DISCLAW_AGENT_NAME and respects the blocklist.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
