Script dev

/script-dev "<description>"

Dev-only TDD pipeline for vallorcine script work. Follows the same test-first discipline as /feature-quick but specialized for the multi-language hook/script pattern used throughout this repo.

Use this when:

• Adding or modifying scripts in scripts/
• Adding or modifying hooks (statusline, stop hook, subagent hook, etc.)
• Changes that need bash + Python + Node.js implementations
• Changes to install.sh, MANIFEST, or settings.json hook registration

Use /ideate instead when:

• The work is broader than scripts (skills, agents, rules, etc.)
• You're designing something new that needs discussion first

---

Step 1 — Understand the change

Read the description and determine:

1. Which scripts are affected? List every file in scripts/ that will change.
2. Is this a new script or a modification? New scripts need all three implementations + wrapper + MANIFEST + install.sh entries.
3. Which test file covers this? Usually tests/scenario-*.sh or tests/test-install.sh.

Read the affected files and their test files before proceeding.

Display a brief summary:

SCRIPT DEV — <description>
─────────────────────────────────────────────
Files:     <list of scripts to modify/create>
Tests:     <test file(s)>
Languages: bash / python / node  (or subset)
─────────────────────────────────────────────
  Type **yes**  to proceed  ·  or: adjust

---

Step 2 — Write tests first

This step is mandatory. No implementation before tests.

2a — Identify test scenarios

For each script change, identify:

• Happy path (normal operation)
• Edge cases (empty input, missing files, malformed state)
• Backwards compatibility (if changing state file formats)
• Cross-implementation parity (if bash/python/node must produce identical output)
• Error handling (permission errors, missing directories, corrupt data)

2b — Write the tests

Add test cases to the appropriate tests/scenario-*.sh file. Follow the existing pattern:

# ── Test N: <description> ─────────────────────────────────────────
echo ""
echo "  testing: <category>"

# Setup
<prepare state>

# Execute
(cd "$PROJECT_PATH" && <run script>)

# Assert
if <condition>; then
    pass "<description>"
else
    fail "<description>" "<diagnostic>"
fi

Test requirements:

• Use /tmp/vallorcine/* paths (pre-granted permissions)
• Test all three language implementations if applicable
• Test malformed/corrupt input gracefully handled (exit 0, no crash)
• Test backwards compatibility with legacy shell variable format if touching state files

2c — Verify tests fail

Run the test file and confirm new tests fail (they test behavior that doesn't exist yet):

bash tests/<test-file>.sh

Show the failure output. If tests pass before implementation, they're not testing the right thing — rewrite them.

Display:

TESTS WRITTEN — <N> new tests, all failing as expected
─────────────────────────────────────────────
  Type **yes**  to proceed to implementation  ·  or: adjust tests

---

Step 3 — Implement

Multi-language implementation order

Always implement in this order:

1. Bash first — the fallback that must always work. No jq required for flat JSON objects (use grep -o / sed patterns). Shell variable backwards-compat via _is_json detection.

2. Python second — stdlib only (json, sys, os, pathlib, datetime). No pip dependencies. Must produce identical side effects (state files, log entries, ANSI output) as bash.

3. Node.js third — stdlib only (fs, path, os, process). No npm dependencies. Must produce identical side effects as bash and Python.

Script conventions

State files: JSON format. Flat objects only — no nesting beyond one level. Bash reads via grep -o '"key":"[^"]*"' | cut -d'"' -f4. Always detect legacy shell variable format with _is_json and read it if found.

Atomic writes: Write to .tmp then rename() / mv. Prevents partial reads by concurrent hooks.

ANSI output: Use \\033[NNm (double-escaped) in Python/Node output strings. The wrapper scripts call echo -e which interprets them. This matches how bash echo -e works natively.

Error handling: All scripts must exit 0 on any error. Hooks that crash with nonzero exit may block Claude Code. Wrap filesystem operations in try/catch (Python: try/except OSError, Node: try {} catch {}).

Wrappers: Every user-facing script has a wrapper at scripts/<name>-wrapper.sh that detects runtimes and delegates: python3 → node → bash. Wrappers catch errors from enhanced implementations and fall back silently.

Implementation checklist

For each script:

• Bash implementation works without jq (for JSON state files)
• Python implementation uses only stdlib
• Node.js implementation uses only stdlib
• All three produce identical side effects (files written, content format)
• All three exit 0 on any error condition
• Atomic writes for state files (.tmp + rename)
• Backwards-compat: reads legacy shell variable format if detected
• Wrapper delegates correctly

Run tests after each implementation

bash tests/<test-file>.sh

Fix failures before moving to the next language.

---

Step 4 — Review

After all tests pass, perform a self-review against these criteria:

Performance review

Scripts run on every assistant message (statusline) or every response (stop hook). They must be fast.

• No unnecessary file reads — don't read a file twice when one read suffices (e.g., return line count from usage parsing, don't re-read)
• No unnecessary stat calls — existsSync before readFileSync is redundant if the read has a try/catch
• Use max() not sorted() when only the most recent file is needed
• Avoid spawning subprocesses — no child_process.exec in Node, no subprocess.run in Python. Pure stdlib file I/O only.
• Target: <10ms for statusline, <5ms for no-op stop hook path

Memory and resource review

• No unbounded reads — don't slurp entire large files into memory when line-by-line streaming works (Python: iterate file object, Node: be aware that readFileSync loads the full file)
• No accumulation — scripts run and exit, no long-lived state. But within a run, don't build arrays of all transcript entries when only a sum is needed.
• Close file handles — use with in Python, don't hold open FDs

Correctness review

• Race conditions — state file reads/writes across concurrent hooks. Atomic writes mitigate but don't eliminate. Document any remaining windows.
• Path assumptions — hooks run from project root (cwd). Don't assume subdirectory or absolute paths unless reading from state.
• Integer arithmetic — parseInt always with radix 10 in Node. Token counts can exceed 32-bit — use standard Number (safe to 2^53).

Distribution review (if new files added)

• MANIFEST updated with all new script paths
• install.sh copies new files and registers any new hooks
• settings.json uses wrapper scripts, not direct implementations
• .gitignore includes any new runtime state files
• tests/test-install.sh asserts new files are installed

---

Step 5 — Final verification

Run the full test suite:

bash tests/scenario-token-tracking.sh && bash tests/test-install.sh

All tests must pass. Display the final count.

SCRIPT DEV COMPLETE
─────────────────────────────────────────────
Tests:  <N> passed, 0 failed
Files:  <list of files changed/created>
─────────────────────────────────────────────