Problem:
- Sync branch setup occurred before the config file was initialized
- Persistence only targeted the database, leading to loss on re-init
Solution:
- Reorder initialization to create the config file before sync setup
- Synchronize sync branch state to both config file and database
Impact:
- Settings are preserved across re-initialization and DB clears
- Better consistency between file and database state
Fixes: #927 (Bug 3)
Replace tool-specific setup commands with a generic recipe-based system.
New tools become config entries, not code changes.
Changes:
- Add internal/recipes/ package with Recipe type and built-in recipes
- Add --list flag to show available recipes
- Add --print flag to output template to stdout
- Add -o flag to write template to arbitrary path
- Add --add flag to save custom recipes to .beads/recipes.toml
- Add built-in recipes: windsurf, cody, kilocode (new)
- Legacy recipes (cursor, claude, gemini, aider, factory) continue to work
The recipe system enables:
- Adding new tool support without code changes
- User-defined recipes in .beads/recipes.toml
- Shared template across all file-based integrations
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
The external_ref field was stored correctly and visible in --json
output, but missing from the human-readable text display.
Added External Ref line after other metadata fields in both daemon
and direct mode paths. Added tests for external_ref display.
The RPC server's handleReady() was explicitly setting Status to
StatusOpen, which overrode the intended behavior where an empty Status
field matches both 'open' and 'in_progress' issues.
Removed the Status field assignment so it remains empty (zero value),
allowing the SQLite storage layer to correctly return both statuses
as documented in the help text.
Fixes #5aml
The --blocks flag handler duplicated cycle detection warning logic from
depAddCmd. Extract to a shared helper function.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The bug: initializeNoDbMode() was setting the legacy global storeActive
but not cmdCtx.StoreActive. When ensureStoreActive() checked
isStoreActive(), it used cmdCtx.StoreActive (which was false), causing
the JSONL-only mode error even when --no-db was passed.
The fix: Use accessor functions (lockStore, setStore, setStoreActive,
unlockStore) which set both the legacy globals and cmdCtx fields.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add alreadyExported flag to skip redundant export. When
gitHasUncommittedBeadsChanges() detects uncommitted changes, we export
at line 175. The flag prevents the normal flow from exporting again
at line 293.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Adds a --blocks (-b) shorthand flag to bd dep command for natural dependency syntax:
bd dep bd-xyz --blocks bd-abc # bd-xyz blocks bd-abc
Equivalent to: bd dep add bd-abc bd-xyz
- Full daemon and direct mode support
- Cycle detection and child-parent anti-pattern checks
- JSON output support
Contributed by: kraitsura
The daemon code path was returning early after adding the dependency,
skipping the cycle detection that runs for direct mode. Restructure
so both paths share the cycle detection and output code.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Agents naturally try to use 'bd dep <blocker> --blocks <blocked>' when
establishing blocking relationships - a desire path revealing intuitive
mental model for how dependencies should work.
When AI agents set up dependency chains, they consistently attempt:
bd dep conduit-abc --blocks conduit-xyz
This reveals a desire path - the syntax users naturally reach for before
reading documentation. Instead of fighting this intuition, we embrace it.
- Add --blocks (-b) flag to the bd dep command
- Support syntax: bd dep <blocker-id> --blocks <blocked-id>
- Equivalent to: bd dep add <blocked-id> <blocker-id>
- Full daemon and direct mode support
- Cycle detection and child-parent anti-pattern checks
- JSON output support for programmatic use
This is purely additive. The existing command structure remains:
- 'bd dep add' subcommand works exactly as before
- All other dep subcommands (remove, list, tree, cycles) unchanged
- No breaking changes to existing workflows
bd dep bd-xyz --blocks bd-abc # bd-xyz blocks bd-abc
bd dep bd-xyz -b bd-abc # Same, using shorthand
bd dep add bd-abc bd-xyz # Original syntax still works
- Added TestDepBlocksFlag for flag initialization
- Added TestDepBlocksFlagFunctionality for semantic correctness
- All existing tests pass
In performExport, if git commit succeeded but push failed, the
finalizeExportMetadata() was never called because we returned early.
This meant metadata would not reflect the successful export+commit.
Now finalize is called:
- Right after syncBranchCommitAndPush succeeds
- Right after gitCommit succeeds (before push attempt)
- When no git changes exist (export still happened)
Push failure still returns early, but metadata is already updated.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
On macOS with case-insensitive filesystem, path casing differences between
the daemon socket path and git worktree registry caused sync failures.
Changed isValidWorktree() to use utils.PathsEqual() which handles
case-insensitivity on macOS/Windows, matching the fix already applied
to daemon registry in GH#869.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When running bd init in a subdirectory of a hub (e.g., ~/Repos/project
where ~/Repos/.beads exists), the new database was incorrectly inheriting
issues from the parent hub.
Root cause: checkGitForIssues() computed the relative path from gitRoot
to beadsDir but did not validate that beadsDir was actually inside the
git repository. When beadsDir was outside (e.g., ../.beads), it would
still attempt to import, causing contamination.
Fix: Add a guard to reject beadsDir paths that start with .. (outside
the git repository boundary).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When setting a slot to a bead from a different beads database
(e.g., setting an hq-* role bead on a gt-* agent bead), the command
now uses prefix-based routing via routes.jsonl to resolve the bead
in the correct database.
Previously, bd slot set only looked in the local database, failing
to find cross-db references like hq-polecat-role from rig beads.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This fixes the atomicity gap where exportToJSONL would update SQLite
metadata (clear dirty flags, update content hash, last_import_time)
BEFORE the git commit. If git commit failed, SQLite would incorrectly
indicate the sync succeeded.
Changes:
- Add ExportResult struct to capture export metadata for deferred finalization
- Add exportToJSONLDeferred() that exports without updating metadata
- Add finalizeExport() to update metadata after git commit succeeds
- Update daemon_sync.go sync flows to defer metadata updates
Now the sync flow is truly atomic: metadata only updates after git commit.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
Add CheckSyncDivergence doctor check that detects:
- JSONL on disk differs from git HEAD version
- SQLite last_import_time does not match JSONL mtime
- Uncommitted .beads/ changes exist
Each issue includes auto-fix suggestions (bd sync, bd export, git commit).
Multiple divergence issues result in error status.
Part of GH#885 recovery mechanism.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add pre-flight safety check to detect when a previous sync exported
but failed before commit, leaving JSONL in an inconsistent state.
- Add gitHasUncommittedBeadsChanges() helper in sync_git.go
- Call in sync pre-flight checks after merge/rebase check
- If uncommitted changes detected, force re-export to reconcile state
This catches the failure mode early before it compounds across worktrees.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
bd sync fails with exit status 128 when the daemon is started from a
terminal with different path casing than what git has stored. This
happens on macOS case-insensitive filesystem when directory names
are renamed (e.g., MyProject to myproject) but terminal sessions
retain the old casing.
The fix uses realpath(1) on macOS to get the true filesystem case
when canonicalizing paths:
- CanonicalizePath() now calls realpath on macOS
- git.GetRepoRoot() canonicalizes repoRoot via canonicalizeCase()
- syncbranch.GetRepoRoot() uses utils.CanonicalizePath()
This ensures git worktree paths match exactly, preventing the
exit status 128 errors from git operations.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
Use git sparse-checkout command instead of manually setting
core.sparseCheckout config. The sparse-checkout command properly
scopes the setting to the worktree via extensions.worktreeConfig,
avoiding the confusing sparse checkout message in git status.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
On macOS (HFS+/APFS), `bd sync` would fail with exit status 128 when
the daemon was started from a terminal session with different path
casing than what git had stored for the worktree (e.g., /Users/.../
MyProject vs /Users/.../myproject).
Fixed by normalizing workspace paths using `filepath.EvalSymlinks()`
before storing in the registry and comparing during lookups:
- registry.Register(): Canonicalizes workspace path before storing
- registry.Unregister(): Canonicalizes paths before comparison
- FindDaemonByWorkspace(): Canonicalizes paths before lookup
This ensures consistent path matching across case-insensitive
filesystems since EvalSymlinks returns the actual filesystem casing.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When bd sync fails mid-operation, the local JSONL can become stale while
the SQLite database has the correct state. Previously, bd doctor only
checked count and timestamp differences, missing cases where counts match
but issue statuses differ.
This adds content-level comparison to CheckDatabaseJSONLSync that:
- Compares issue statuses between DB and JSONL
- Samples up to 500 issues for performance on large databases
- Reports detailed mismatches (shows up to 3 examples)
- Suggests 'bd export' to fix the stale JSONL
Example detection:
Status mismatch: 1 issue(s) have different status in DB vs JSONL
Status mismatches detected:
test-1: DB=closed, JSONL=open
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add flag-based alternatives to the positional argument for `bd dep add`:
- `--blocked-by <id>`: Specify the blocking issue via flag
- `--depends-on <id>`: Alias for --blocked-by
This reduces token waste when Claude guesses flag-based syntax, which
is a common pattern. Previously, Claude would attempt commands like:
bd dep add issue-123 --blocked-by issue-456
This would fail with "unknown flag" and require retry. Now both:
bd dep add issue-123 issue-456
bd dep add issue-123 --blocked-by issue-456
work identically.
Closes GH#888
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The close_reason and closed_by_session fields were being silently dropped
during 3-way merge operations because the simplified Issue struct in
internal/merge/merge.go was missing these fields.
Changes:
- Add CloseReason and ClosedBySession fields to merge.Issue struct
- Implement merge logic that preserves these fields when status is closed
- Use timestamp-based conflict resolution (later closed_at wins)
- Clear close metadata when status becomes non-closed
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
On Windows, the downloaded ZIP file may remain locked by antivirus
software or Node.js file handle release delays. This causes
Expand-Archive to fail with "being used by another process" errors.
Added exponential backoff retry logic (5 attempts, 500ms-8s delays)
that detects file lock errors and retries the extraction. Non-lock
errors still fail immediately.
Fixes GH#889 (bd-5dlz)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Git 2.38+ enables core.sparseCheckout on the main repo as a side effect
of worktree creation, causing confusing git status message:
"You are in a sparse checkout with 100% of tracked files present."
The fix explicitly disables sparse checkout on the main repo after
creating the beads worktree. This doesn't affect the worktree's sparse
checkout functionality since the patterns are already applied during
checkout.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
When gitRefsPath is empty (not in a git repo), strings.HasPrefix(path, "")
always returns true, causing every file write in .beads/ directory
(including daemon.log) to trigger debouncer and create an event storm.
This fix adds a check to ensure gitRefsPath is not empty before the
HasPrefix comparison.
Fixes the issue where daemon.log grows rapidly (17MB+) due to the
self-triggering loop: write log -> detect change -> write log -> ...
Co-authored-by: Test User <test@example.com>
When not in a git repository:
- Daemon startup now shows clear message immediately instead of waiting
5 seconds: "Note: No git repository initialized — running without
background sync"
- Added new doctor check "Git Sync Setup" that explains the situation
Doctor check now shows three states:
1. No git repo → Warning with fix: "Run 'git init' to enable background sync"
2. Git repo, no sync-branch → OK with hint about team collaboration benefits
3. Git repo + sync-branch → OK, fully configured
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
feat: add multi-prefix support via allowed_prefixes config
Adds support for allowing multiple prefixes in a beads database via the
allowed_prefixes config key. This is needed for Gas Town which uses different
prefixes for different purposes (hq-, gt-, rig-specific prefixes).
Usage: bd config set allowed_prefixes "gt,hq,hmc"
PR #881 by @web3dev1337
- Pull from sync-branch before rename if configured
- Import all issues from JSONL before rename to prevent data loss
- Export directly to JSONL after rename (don't rely on flushManager)
- Apply same pattern to --repair mode
- Add newSilentLogger() for production use (not test-only)
- Add comprehensive tests for JSONL update scenarios
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Allow team-wide auto-sync configuration via config.yaml instead of SQLite.
This enables teams to share auto-commit/auto-push settings through version control.
Changes:
- Add daemon.auto_commit, daemon.auto_push, daemon.auto_pull to YamlOnlyKeys
- Add daemon.* prefix to YAML-only prefixes
- Update daemon startup to read from config.yaml first, then fall back to SQLite
- Update bd init --team to write daemon settings to config.yaml
Usage:
# In .beads/config.yaml (version controlled, shared by team)
daemon.auto_commit: true
daemon.auto_push: true
# Or via bd config set
bd config set daemon.auto_commit true
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Users can now place a .beads/PRIME.md file to fully customize the
workflow instructions output by `bd prime`. The --export flag outputs
the default content for use as a starting template.
Local .beads/PRIME.md is checked first, then redirected location,
allowing clone-specific customization even with shared beads.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Review findings addressed:
1. Fixed HasSyncBranchGitignoreFlags() - now correctly returns (hasAnyFlag,
hasSkipWorktree) since skip-worktree takes precedence in git ls-files -v
2. Added interactions.jsonl to list of files to hide (was only issues.jsonl)
3. Added idempotency check - skips setting flags if already set (checks for S)
4. Made output conditional - only prints when flags actually changed
5. Fixed addToGitExclude() pattern matching - now uses exact line match
instead of substring to prevent false positives
6. Refactored to use setGitIndexFlags() helper to reduce duplication
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
- Guard deploy-docs job to only run in canonical repository
- Guard goreleaser job to only run in canonical repository
- Guard update-homebrew job to only run in canonical repository
- Guard test-pypi job to only run in canonical repository
Prevents fork workflows from attempting to deploy, release, or publish to external services.
When sync.branch is configured, issues.jsonl appears modified in git status
even though changes go to the sync branch. This is confusing for users and
risks accidental commits to the wrong branch.
Implementation:
- Added SyncBranchGitignore() to set git index flags (assume-unchanged,
skip-worktree) on issues.jsonl when sync.branch is configured
- For untracked files, adds to .git/info/exclude instead
- Called automatically from bd sync after successful sync-branch sync
- Added bd doctor check and fix for this issue
- Added HasSyncBranchGitignoreFlags() to check current flag state
- Added ClearSyncBranchGitignore() to remove flags when sync.branch disabled
Fixes: GH#870 (duplicate of GH#797, GH#801)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
The --protect-left-snapshot mechanism was protecting ALL local issues
by ID alone, ignoring timestamps. This caused newer remote changes to
be incorrectly skipped during cross-worktree sync.
Changes:
- Add BuildIDToTimestampMap() to SnapshotManager for timestamp-aware
snapshot reading
- Change ProtectLocalExportIDs from map[string]bool to map[string]time.Time
- Add shouldProtectFromUpdate() helper that compares timestamps
- Only protect if local snapshot is newer than incoming; allow update
if incoming is newer
This fixes data loss scenarios where:
1. Main worktree closes issue at 11:31
2. Test worktree syncs and incorrectly skips the update
3. Test worktree then pushes stale open state, overwriting mains changes
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Complements the validation fix from c6fe9d71 - the parseAgentIDFields
function now also scans right-to-left for known role tokens instead
of using fixed position parsing.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The sync.remote config was being set via `bd config set sync.remote <name>`
but `bd sync` was still using 'origin' for git pull/push operations.
Changes:
- Updated gitPull() and gitPush() in sync_git.go to accept a configuredRemote
parameter that takes precedence over git's branch tracking config
- Updated sync.go to read sync.remote config and pass it to gitPull/gitPush
- Updated daemon_sync.go to read sync.remote config for all daemon sync ops
- Added user-facing messages to show which remote is being used
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Executed-By: beads/crew/dave
Rig: beads
Role: crew
On macOS and Windows, filesystems are typically case-insensitive, so
/Users/foo/Desktop and /Users/foo/desktop refer to the same directory.
The daemon registry and discovery code was doing direct string comparison,
causing path mismatches when the casing differed.
Fix:
- Add NormalizePathForComparison() and PathsEqual() to internal/utils/path.go
- These resolve symlinks and lowercase paths on darwin/windows
- Update all workspace path comparisons in registry.go, discovery.go, and
daemons.go to use PathsEqual()
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Ensures AwaitType, AwaitID, and Timeout fields are carried over when
cloning template subgraphs, enabling async coordination in instantiated
molecules.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Rewrites ValidateAgentID to scan right-to-left for known role tokens
instead of relying on fixed position parsing. This allows rig names
containing hyphens (e.g., ob-my-project-witness).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
The bd agent state, heartbeat, and show commands now respect
routes.jsonl for cross-repo lookups, matching the behavior of
bd show.
Previously, these commands used utils.ResolvePartialID directly,
which bypassed routing. Now they use resolveAndGetIssueWithRouting
and needsRouting checks, consistent with show.go.
GetYamlConfig was not normalizing key aliases (e.g., sync.branch ->
sync-branch), causing 'bd config get sync.branch' to return 'not set'
even when the value was correctly stored.
SetYamlConfig already normalized keys, but GetYamlConfig did not,
leading to a confusing mismatch where set appeared to work but get
could not find the value.
Added TestGetYamlConfig_KeyNormalization to verify the fix.
Fixes#873
Adds a new slot bead type and merge-slot commands for serializing
conflict resolution in the merge queue. This prevents "monkey knife
fights" where multiple polecats race to resolve conflicts.
- Add TypeSlot to bead types
- Add Holder field to Issue struct
- Add bd merge-slot create/check/acquire/release commands
- Add Holder field to UpdateArgs in RPC protocol
(gt-4u49x)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Beads only had the skill version (Claude-only). This adds the command
version so users can type /handoff directly like in gastown.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add BenchmarkRebuildBlockedCache_Large and _XLarge to measure cache
rebuild performance in isolation.
Results show cache rebuild takes ~773ms at 10K issues and ~1.3s at 20K,
significantly slower than the ~50ms originally estimated in the issue.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
