dirsql 0.0.15__tar.gz → 0.0.17__tar.gz

dirsql-0.0.17/.claude/CLAUDE.md

@@ -0,0 +1,3 @@
+# dirsql
+
+All project conventions, workflow rules, and development instructions are in [AGENTS.md](../AGENTS.md).

{dirsql-0.0.15 → dirsql-0.0.17}/.gitignore

@@ -1,7 +1,6 @@
 /target
 .beads
 .claude/settings.local.json
-AGENTS.md
 
 # Dolt database files (added by bd init)
 .dolt/

{dirsql-0.0.15 → dirsql-0.0.17}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dirsql
-Version: 0.0.15
+Version: 0.0.17
 Requires-Dist: pytest>=8 ; extra == 'dev'
 Requires-Dist: pytest-describe>=2 ; extra == 'dev'
 Requires-Dist: pytest-asyncio>=0.23 ; extra == 'dev'

{dirsql-0.0.15 → dirsql-0.0.17}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "maturin"
 
 [project]
 name = "dirsql"
-version = "0.0.15"
+version = "0.0.17"
 description = "Ephemeral SQL index over a local directory"
 license = "MIT"
 requires-python = ">=3.12"

dirsql-0.0.15/.claude/CLAUDE.md

@@ -1,122 +0,0 @@
-# dirsql Development
-
-## Scratch Files
-
-Write scratch/temporary files to `/tmp` instead of asking permission. Use unique filenames to avoid collisions with other sessions.
-
-## Workflow
-
-- Work in git worktrees under `.worktrees/` folder
-- **NEVER commit directly to main** - always create a PR
-- One PR per bead. Beads should be concise and small -- as small as possible while still being useful
-- Use `bd` (Beads) for task tracking: `bd list`, `bd show <id>`, `bd ready`
-- **Bead first**: When starting new work, the first step is always to create a bead (`bd create`). No implementation work begins without a bead.
-
-### Git Worktrees
-
-**ALL work happens in git worktrees.** Never edit files in the root repo directory. Never commit outside a worktree.
-
-#### Creating a Worktree
-
-```bash
-git worktree add .worktrees/my-feature -b feat/my-feature
-cd .worktrees/my-feature
-```
-
-#### Removing a Worktree
-
-**DANGER: removing a worktree while your shell CWD is inside it permanently breaks the shell.** The ONLY safe procedure:
-
-```bash
-# Step 1: Move CWD to the root repo FIRST (not optional)
-cd /home/duncan/work/code/projects/dirsql
-
-# Step 2: Now remove the worktree
-git worktree remove .worktrees/my-feature
-```
-
-**Do NOT skip step 1. Do NOT substitute `git -C` for `cd`.**
-
-### Beads Workflow
-
-**Lifecycle:**
-1. **Claim it FIRST**: `bd update <id> --claim` before any work
-2. **Create worktree and branch**
-3. **Link the PR**: `bd update <id> --external-ref "gh-<pr-number>"` after creating the PR
-4. **Close**: `bd close <id>` immediately after the PR is merged
-
-### Subagent Workflow
-
-New work on beads should be done via subagents in isolated worktrees. Each subagent:
-1. Claims the bead (`bd update <id> --claim`) before starting any work
-2. Creates a worktree and branch for its bead
-3. Does the implementation work (red/green TDD)
-4. Pushes the branch and opens a PR
-5. Monitors the PR and proactively resolves:
-   - CI failures
-   - GPG signing complaints
-   - Merge conflicts
-6. Continues monitoring until the PR is in a mergeable state
-
-### Orchestrator Responsibilities
-
-The orchestrator (main Claude session) must proactively:
-1. **Monitor all open PRs** -- don't wait for the user to report failures. Check CI status after agent completion and on an ongoing basis.
-2. **Fix CI failures** on open PRs immediately, either directly or by dispatching a fix agent.
-3. **Handle post-merge cleanup** as soon as a PR merges (pull main, remove worktree, delete branch, close bead).
-4. **Keep the user informed** of PR status without being asked.
-5. **Use foreground monitoring** when waiting on CI and there's no other work to do. Background monitoring causes the conversation to go silent -- use it only when there's genuinely parallel work to perform.
-6. **Scripts to `/tmp`**: For polling/monitoring scripts (watching CI, waiting for merges), write the script to `/tmp` then run it via `bash /tmp/script.sh`. Do not use inline bash loops in tool calls.
-
-### Post-Merge Cleanup
-
-After a PR merges, the agent (or orchestrator) must:
-1. Pull main in the **root repo**: `git -C /home/duncan/work/code/projects/dirsql pull origin main`
-2. **Move CWD to root repo first** (CRITICAL -- never remove a worktree from inside it): `cd /home/duncan/work/code/projects/dirsql`
-3. Remove the worktree: `git worktree remove .worktrees/<name>`
-4. Delete the local branch: `git branch -d <branch-name>`
-5. **Verify the bead is addressed** by the merged PR, then close it: `bd close <id>`
-
-## Testing
-
-### Red/Green Development
-
-Follow **red/green** (test-first) methodology:
-
-1. **Write the test first** -- it must capture the desired behavior
-2. **Run it and confirm it fails (RED)** -- do NOT proceed until the test turns red reliably. A test that passes before implementation proves nothing.
-3. **Make the minimal change to pass (GREEN)** -- only then write the implementation
-4. Refactor if needed, keeping tests green
-
-### TDD Order: Outside-In
-
-Tests are written **before** implementation, starting from the outermost layer:
-
-1. **Integration test first** -- proves the feature works from the consumer's perspective
-2. **Unit tests** -- written as you implement each module
-
-A feature is not done until integration tests pass and cover the new functionality.
-
-### When to Write What
-
-**Does the commit change the public-facing API?**
-- Yes -> **integration test required**, plus unit tests as you go
-- No -> Check if adequate integration coverage already exists:
-  - Adequate -> unit tests only
-  - Gaps -> add the missing integration tests, plus unit tests
-
-**Always write unit tests.** The question is whether you also need integration tests.
-
-### Test Locations
-
-- **Unit tests**: Colocated with source
-  - Python: `foo.py` -> `foo_test.py` in same directory
-  - Rust: inline `#[cfg(test)]` module at bottom of each source file
-- **Integration tests**: `tests/integration/` -- test the Python SDK layer, mock third-party deps (SQLite, LLM calls). Heavy use of pytest fixtures. Run in CI.
-- **E2E tests**: `tests/e2e/` -- real filesystem, real SQLite, real LLM calls, no mocks. Heavy use of pytest fixtures. **NOT run in CI** (eventual LLM calls make them non-free). Run locally by Claude after significant code changes.
-
-### E2E Test Policy
-
-E2E tests are your primary feedback mechanism. Run them liberally after significant changes -- they catch issues that integration tests miss because integration tests mock out SQLite and (eventually) LLM calls. But do NOT add them to CI workflows. They are a local development tool.
-
-See skillet or karat for examples of test organization, fixtures, and pytest-describe patterns.