@caddis/cli 0.2.6 → 0.4.0

package/bundles/codex/.codex/skills/workflow/skill-creator/scripts/run_loop.py

@@ -0,0 +1,328 @@
+#!/usr/bin/env python3
+"""Run the eval + improve loop until all pass or max iterations reached.
+
+Combines run_eval.py and improve_description.py in a loop, tracking history
+and returning the best description found. Supports train/test split to prevent
+overfitting.
+"""
+
+import argparse
+import json
+import random
+import sys
+import tempfile
+import time
+import webbrowser
+from pathlib import Path
+
+from scripts.generate_report import generate_html
+from scripts.improve_description import improve_description
+from scripts.run_eval import find_project_root, run_eval
+from scripts.utils import parse_skill_md
+
+
+def split_eval_set(eval_set: list[dict], holdout: float, seed: int = 42) -> tuple[list[dict], list[dict]]:
+    """Split eval set into train and test sets, stratified by should_trigger."""
+    random.seed(seed)
+
+    # Separate by should_trigger
+    trigger = [e for e in eval_set if e["should_trigger"]]
+    no_trigger = [e for e in eval_set if not e["should_trigger"]]
+
+    # Shuffle each group
+    random.shuffle(trigger)
+    random.shuffle(no_trigger)
+
+    # Calculate split points
+    n_trigger_test = max(1, int(len(trigger) * holdout))
+    n_no_trigger_test = max(1, int(len(no_trigger) * holdout))
+
+    # Split
+    test_set = trigger[:n_trigger_test] + no_trigger[:n_no_trigger_test]
+    train_set = trigger[n_trigger_test:] + no_trigger[n_no_trigger_test:]
+
+    return train_set, test_set
+
+
+def run_loop(
+    eval_set: list[dict],
+    skill_path: Path,
+    description_override: str | None,
+    num_workers: int,
+    timeout: int,
+    max_iterations: int,
+    runs_per_query: int,
+    trigger_threshold: float,
+    holdout: float,
+    model: str,
+    verbose: bool,
+    live_report_path: Path | None = None,
+    log_dir: Path | None = None,
+) -> dict:
+    """Run the eval + improvement loop."""
+    project_root = find_project_root()
+    name, original_description, content = parse_skill_md(skill_path)
+    current_description = description_override or original_description
+
+    # Split into train/test if holdout > 0
+    if holdout > 0:
+        train_set, test_set = split_eval_set(eval_set, holdout)
+        if verbose:
+            print(f"Split: {len(train_set)} train, {len(test_set)} test (holdout={holdout})", file=sys.stderr)
+    else:
+        train_set = eval_set
+        test_set = []
+
+    history = []
+    exit_reason = "unknown"
+
+    for iteration in range(1, max_iterations + 1):
+        if verbose:
+            print(f"\n{'='*60}", file=sys.stderr)
+            print(f"Iteration {iteration}/{max_iterations}", file=sys.stderr)
+            print(f"Description: {current_description}", file=sys.stderr)
+            print(f"{'='*60}", file=sys.stderr)
+
+        # Evaluate train + test together in one batch for parallelism
+        all_queries = train_set + test_set
+        t0 = time.time()
+        all_results = run_eval(
+            eval_set=all_queries,
+            skill_name=name,
+            description=current_description,
+            num_workers=num_workers,
+            timeout=timeout,
+            project_root=project_root,
+            runs_per_query=runs_per_query,
+            trigger_threshold=trigger_threshold,
+            model=model,
+        )
+        eval_elapsed = time.time() - t0
+
+        # Split results back into train/test by matching queries
+        train_queries_set = {q["query"] for q in train_set}
+        train_result_list = [r for r in all_results["results"] if r["query"] in train_queries_set]
+        test_result_list = [r for r in all_results["results"] if r["query"] not in train_queries_set]
+
+        train_passed = sum(1 for r in train_result_list if r["pass"])
+        train_total = len(train_result_list)
+        train_summary = {"passed": train_passed, "failed": train_total - train_passed, "total": train_total}
+        train_results = {"results": train_result_list, "summary": train_summary}
+
+        if test_set:
+            test_passed = sum(1 for r in test_result_list if r["pass"])
+            test_total = len(test_result_list)
+            test_summary = {"passed": test_passed, "failed": test_total - test_passed, "total": test_total}
+            test_results = {"results": test_result_list, "summary": test_summary}
+        else:
+            test_results = None
+            test_summary = None
+
+        history.append({
+            "iteration": iteration,
+            "description": current_description,
+            "train_passed": train_summary["passed"],
+            "train_failed": train_summary["failed"],
+            "train_total": train_summary["total"],
+            "train_results": train_results["results"],
+            "test_passed": test_summary["passed"] if test_summary else None,
+            "test_failed": test_summary["failed"] if test_summary else None,
+            "test_total": test_summary["total"] if test_summary else None,
+            "test_results": test_results["results"] if test_results else None,
+            # For backward compat with report generator
+            "passed": train_summary["passed"],
+            "failed": train_summary["failed"],
+            "total": train_summary["total"],
+            "results": train_results["results"],
+        })
+
+        # Write live report if path provided
+        if live_report_path:
+            partial_output = {
+                "original_description": original_description,
+                "best_description": current_description,
+                "best_score": "in progress",
+                "iterations_run": len(history),
+                "holdout": holdout,
+                "train_size": len(train_set),
+                "test_size": len(test_set),
+                "history": history,
+            }
+            live_report_path.write_text(generate_html(partial_output, auto_refresh=True, skill_name=name))
+
+        if verbose:
+            def print_eval_stats(label, results, elapsed):
+                pos = [r for r in results if r["should_trigger"]]
+                neg = [r for r in results if not r["should_trigger"]]
+                tp = sum(r["triggers"] for r in pos)
+                pos_runs = sum(r["runs"] for r in pos)
+                fn = pos_runs - tp
+                fp = sum(r["triggers"] for r in neg)
+                neg_runs = sum(r["runs"] for r in neg)
+                tn = neg_runs - fp
+                total = tp + tn + fp + fn
+                precision = tp / (tp + fp) if (tp + fp) > 0 else 1.0
+                recall = tp / (tp + fn) if (tp + fn) > 0 else 1.0
+                accuracy = (tp + tn) / total if total > 0 else 0.0
+                print(f"{label}: {tp+tn}/{total} correct, precision={precision:.0%} recall={recall:.0%} accuracy={accuracy:.0%} ({elapsed:.1f}s)", file=sys.stderr)
+                for r in results:
+                    status = "PASS" if r["pass"] else "FAIL"
+                    rate_str = f"{r['triggers']}/{r['runs']}"
+                    print(f"  [{status}] rate={rate_str} expected={r['should_trigger']}: {r['query'][:60]}", file=sys.stderr)
+
+            print_eval_stats("Train", train_results["results"], eval_elapsed)
+            if test_summary:
+                print_eval_stats("Test ", test_results["results"], 0)
+
+        if train_summary["failed"] == 0:
+            exit_reason = f"all_passed (iteration {iteration})"
+            if verbose:
+                print(f"\nAll train queries passed on iteration {iteration}!", file=sys.stderr)
+            break
+
+        if iteration == max_iterations:
+            exit_reason = f"max_iterations ({max_iterations})"
+            if verbose:
+                print(f"\nMax iterations reached ({max_iterations}).", file=sys.stderr)
+            break
+
+        # Improve the description based on train results
+        if verbose:
+            print(f"\nImproving description...", file=sys.stderr)
+
+        t0 = time.time()
+        # Strip test scores from history so improvement model can't see them
+        blinded_history = [
+            {k: v for k, v in h.items() if not k.startswith("test_")}
+            for h in history
+        ]
+        new_description = improve_description(
+            skill_name=name,
+            skill_content=content,
+            current_description=current_description,
+            eval_results=train_results,
+            history=blinded_history,
+            model=model,
+            log_dir=log_dir,
+            iteration=iteration,
+        )
+        improve_elapsed = time.time() - t0
+
+        if verbose:
+            print(f"Proposed ({improve_elapsed:.1f}s): {new_description}", file=sys.stderr)
+
+        current_description = new_description
+
+    # Find the best iteration by TEST score (or train if no test set)
+    if test_set:
+        best = max(history, key=lambda h: h["test_passed"] or 0)
+        best_score = f"{best['test_passed']}/{best['test_total']}"
+    else:
+        best = max(history, key=lambda h: h["train_passed"])
+        best_score = f"{best['train_passed']}/{best['train_total']}"
+
+    if verbose:
+        print(f"\nExit reason: {exit_reason}", file=sys.stderr)
+        print(f"Best score: {best_score} (iteration {best['iteration']})", file=sys.stderr)
+
+    return {
+        "exit_reason": exit_reason,
+        "original_description": original_description,
+        "best_description": best["description"],
+        "best_score": best_score,
+        "best_train_score": f"{best['train_passed']}/{best['train_total']}",
+        "best_test_score": f"{best['test_passed']}/{best['test_total']}" if test_set else None,
+        "final_description": current_description,
+        "iterations_run": len(history),
+        "holdout": holdout,
+        "train_size": len(train_set),
+        "test_size": len(test_set),
+        "history": history,
+    }
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Run eval + improve loop")
+    parser.add_argument("--eval-set", required=True, help="Path to eval set JSON file")
+    parser.add_argument("--skill-path", required=True, help="Path to skill directory")
+    parser.add_argument("--description", default=None, help="Override starting description")
+    parser.add_argument("--num-workers", type=int, default=10, help="Number of parallel workers")
+    parser.add_argument("--timeout", type=int, default=30, help="Timeout per query in seconds")
+    parser.add_argument("--max-iterations", type=int, default=5, help="Max improvement iterations")
+    parser.add_argument("--runs-per-query", type=int, default=3, help="Number of runs per query")
+    parser.add_argument("--trigger-threshold", type=float, default=0.5, help="Trigger rate threshold")
+    parser.add_argument("--holdout", type=float, default=0.4, help="Fraction of eval set to hold out for testing (0 to disable)")
+    parser.add_argument("--model", required=True, help="Model for improvement")
+    parser.add_argument("--verbose", action="store_true", help="Print progress to stderr")
+    parser.add_argument("--report", default="auto", help="Generate HTML report at this path (default: 'auto' for temp file, 'none' to disable)")
+    parser.add_argument("--results-dir", default=None, help="Save all outputs (results.json, report.html, log.txt) to a timestamped subdirectory here")
+    args = parser.parse_args()
+
+    eval_set = json.loads(Path(args.eval_set).read_text())
+    skill_path = Path(args.skill_path)
+
+    if not (skill_path / "SKILL.md").exists():
+        print(f"Error: No SKILL.md found at {skill_path}", file=sys.stderr)
+        sys.exit(1)
+
+    name, _, _ = parse_skill_md(skill_path)
+
+    # Set up live report path
+    if args.report != "none":
+        if args.report == "auto":
+            timestamp = time.strftime("%Y%m%d_%H%M%S")
+            live_report_path = Path(tempfile.gettempdir()) / f"skill_description_report_{skill_path.name}_{timestamp}.html"
+        else:
+            live_report_path = Path(args.report)
+        # Open the report immediately so the user can watch
+        live_report_path.write_text("<html><body><h1>Starting optimization loop...</h1><meta http-equiv='refresh' content='5'></body></html>")
+        webbrowser.open(str(live_report_path))
+    else:
+        live_report_path = None
+
+    # Determine output directory (create before run_loop so logs can be written)
+    if args.results_dir:
+        timestamp = time.strftime("%Y-%m-%d_%H%M%S")
+        results_dir = Path(args.results_dir) / timestamp
+        results_dir.mkdir(parents=True, exist_ok=True)
+    else:
+        results_dir = None
+
+    log_dir = results_dir / "logs" if results_dir else None
+
+    output = run_loop(
+        eval_set=eval_set,
+        skill_path=skill_path,
+        description_override=args.description,
+        num_workers=args.num_workers,
+        timeout=args.timeout,
+        max_iterations=args.max_iterations,
+        runs_per_query=args.runs_per_query,
+        trigger_threshold=args.trigger_threshold,
+        holdout=args.holdout,
+        model=args.model,
+        verbose=args.verbose,
+        live_report_path=live_report_path,
+        log_dir=log_dir,
+    )
+
+    # Save JSON output
+    json_output = json.dumps(output, indent=2)
+    print(json_output)
+    if results_dir:
+        (results_dir / "results.json").write_text(json_output)
+
+    # Write final HTML report (without auto-refresh)
+    if live_report_path:
+        live_report_path.write_text(generate_html(output, auto_refresh=False, skill_name=name))
+        print(f"\nReport: {live_report_path}", file=sys.stderr)
+
+    if results_dir and live_report_path:
+        (results_dir / "report.html").write_text(generate_html(output, auto_refresh=False, skill_name=name))
+
+    if results_dir:
+        print(f"Results saved to: {results_dir}", file=sys.stderr)
+
+
+if __name__ == "__main__":
+    main()

package/bundles/codex/.codex/skills/workflow/skill-creator/scripts/utils.py

@@ -0,0 +1,47 @@
+"""Shared utilities for skill-creator scripts."""
+
+from pathlib import Path
+
+
+
+def parse_skill_md(skill_path: Path) -> tuple[str, str, str]:
+    """Parse a SKILL.md file, returning (name, description, full_content)."""
+    content = (skill_path / "SKILL.md").read_text()
+    lines = content.split("\n")
+
+    if lines[0].strip() != "---":
+        raise ValueError("SKILL.md missing frontmatter (no opening ---)")
+
+    end_idx = None
+    for i, line in enumerate(lines[1:], start=1):
+        if line.strip() == "---":
+            end_idx = i
+            break
+
+    if end_idx is None:
+        raise ValueError("SKILL.md missing frontmatter (no closing ---)")
+
+    name = ""
+    description = ""
+    frontmatter_lines = lines[1:end_idx]
+    i = 0
+    while i < len(frontmatter_lines):
+        line = frontmatter_lines[i]
+        if line.startswith("name:"):
+            name = line[len("name:"):].strip().strip('"').strip("'")
+        elif line.startswith("description:"):
+            value = line[len("description:"):].strip()
+            # Handle YAML multiline indicators (>, |, >-, |-)
+            if value in (">", "|", ">-", "|-"):
+                continuation_lines: list[str] = []
+                i += 1
+                while i < len(frontmatter_lines) and (frontmatter_lines[i].startswith("  ") or frontmatter_lines[i].startswith("\t")):
+                    continuation_lines.append(frontmatter_lines[i].strip())
+                    i += 1
+                description = " ".join(continuation_lines)
+                continue
+            else:
+                description = value.strip('"').strip("'")
+        i += 1
+
+    return name, description, content

package/bundles/codex/AGENTS.md

@@ -0,0 +1,146 @@
+# {{PROJECT_NAME}} — Project Memory (canonical rules)
+
+> **This is the canonical rules file.** Every coding agent reads it: Claude Code (via a one-line
+> `@AGENTS.md` import in `CLAUDE.md`), OpenAI Codex CLI, and Antigravity (`agy`) all load `AGENTS.md`
+> directly. `CLAUDE.md` is a thin shim — it imports this file and adds only Claude-native
+> conveniences. **The laws and the harness are identical across agents;** only native conveniences
+> differ. Keep durable project rules HERE, never in the shim.
+
+## What this project is
+{{PROJECT_DESCRIPTION}}
+
+**Stack:** {{STACK_SUMMARY}}
+{{STACK_REFERENCE_LINE}}
+
+## The Laws (non-negotiable)
+1. **TDD always.** No production code without a failing test first. Red → Green → Refactor. A behavior
+   change must start with a failing test. Never write the test after the code and call it TDD.
+2. **No absolute paths.** Resolve from file/project root. Portable always.
+3. **No silent failures.** Specific exceptions, logged. Never swallow errors without a trace.
+4. **Verify, don't assume.** Run the build/lint/test before claiming done. State what you ran.
+5. **Honest status.** "Partially done, here's what remains" beats a false "done". Report failures
+   with their output. If a step was skipped, say so.
+6. **Stay in scope.** Do the phase's work and stop at its exit gate. Don't gold-plate.
+
+## The Development Harness (how we work)
+Every non-trivial change follows this loop. The **plan file is the durable spine** — it survives
+session death, so progress is never lost to context limits.
+
+```
+1. PLAN      write a phased plan to .caddis/plans/<feature>.md
+2. per phase:
+   a. RED       write the failing test first   (TDD — see Law 1)
+   b. GREEN     minimal code to pass
+   c. REFACTOR  clean up, tests stay green
+   d. VERIFY    run tests + lint; do heavy reads (audit, review, preflight) as a bounded,
+                summarized step (Claude Code: dispatch a subagent that reports back)
+   e. CHECKPOINT  update the plan's tracker → commit
+3. END       update .caddis/relay.md so the next session resumes (read it first on resume)
+```
+
+### Context discipline (defense against context rot)
+Anything that **reads a lot but writes a little** — codebase audit, review, preflight, test runs,
+security scan — should run as a **bounded step whose only output is a conclusion**, not a raw dump.
+Claude Code has subagents with their own context for exactly this; agents without a separate context
+(Codex, agy) do the heavy read as a distinct step and summarize into the plan or a doc under
+`.caddis/agent-docs/` before continuing. This is the primary defense against session overflow.
+
+### Resuming a session
+On a fresh session read in order: `.caddis/relay.md` (if present) → the active plan in
+`.caddis/plans/` → its tracker. The conversation is disposable; these files are the truth.
+
+## Always-active conventions
+- **Commits**: Conventional Commits (`feat:`, `fix:`, `test:`, `chore:`). One logical change per commit.
+- **Terse**: lead with the answer; results/code over prose.
+- **Secrets**: never commit; keep env/config secrets git-ignored. Never print credentials.
+- **Document frontmatter**: Every descriptive Markdown deliverable you write (plan, PRD, ADR, design doc,
+  runbook, analysis, handoff) must open with a YAML frontmatter block:
+  ```yaml
+  ---
+  type: plan|prd|adr|design|runbook|handoff|analysis|review
+  status: draft|current|done|superseded
+  feature: <feature-slug>
+  creation-agent: caddis
+  Original Author: <agent or human name>
+  Creation Date: <YYYY-MM-DDTHH:MM:SSZ>
+  Creating Model: <exact model ID>
+  ---
+  ```
+  On update, add/refresh: `Last Author`, `Last Updated`, `Last Model Used`.
+  Do NOT add frontmatter to: `relay.md`, `AGENTS.md`, `CLAUDE.md`, skill files, or agent config files.
+
+## Where things live
+| Need | Location |
+|---|---|
+| Reference-doc index (the meta-KB) — **read first** | `.caddis/kb/DOC-MAP.md` |
+| **Environment facts — hosts, logins, repo locations, what a folder really is** | `.caddis/kb/environment-map.md` |
+| Every UI page → endpoints → DB (frontend repos) | `UI_PAGE_GUIDE.md` |
+| Active plans | `.caddis/plans/<feature>.md` |
+| PRDs | `.caddis/prd/<feature-slug>.md` |
+| **Anything to do LATER — defect, proposal, parked plan, idea** | **`.caddis/parking-lot/<slug>.md`** |
+| Outbound asks only a human outside this repo can resolve | `.caddis/comms/` + its `register.md` |
+| Standing prompts / drivers / specs handed to an agent | `.caddis/prompts/` |
+| Review output (code-review, cross-review) | `.caddis/reviews/` |
+| ADRs, design docs, runbooks | `docs/adr/`, `docs/design/`, `docs/runbooks/` (team-facing, not tool scratch) |
+| Agent-produced docs (evals, reviews, debug) | `.caddis/agent-docs/` |
+| Session resume doc | `.caddis/relay.md` |
+| Per-area conventions | `src/AGENTS.md`, `frontend/AGENTS.md`, `tests/AGENTS.md` (as present) |
+
+**`.caddis/` is the default home for working artifacts.** A plan → `.caddis/plans/`, a PRD →
+`.caddis/prd/`, the KB → `.caddis/kb/`, a prompt/driver/spec → `.caddis/prompts/`, a review →
+`.caddis/reviews/`. Interpret an unqualified reference ("write a plan", "save this prompt")
+accordingly — never scatter these to the repo root or `.github/` (the latter is often a published/
+synced tree in harness-authoring repos).
+
+> **A repo created before the rename uses `.claudster/` instead.** Run `/caddis:migrate-dir` to
+> convert it — everything under `.caddis/` above is invisible to caddis until you do.
+
+### Future work goes in ONE place
+**`.caddis/parking-lot/` is the whole backlog. One item, one file.** If it is not there, it is not
+on the backlog. Do **not** leave future work in a "things owed" section of a plan or `relay.md`, in
+a KB note, in a chat reply, or in a new register file — those all look like a record and none of
+them can be counted. File it with `/caddis:park`, or write it by hand with this frontmatter:
+
+```yaml
+type: parking-lot        # always
+status: open             # open | doing | done | dropped  (closed vocabulary)
+future: yes              # OPTIONAL. yes = committed. Absent or `no` = a candidate.
+```
+
+`status` and `future` answer different questions and must stay separate: where the item is in its
+life, and whether we have agreed to do it. `python scripts/caddis_tidy.py --check` and the
+`caddis_gate.py parking-lot` gate both fail on a non-conforming item, so this is enforced, not
+advised. Two neighbours that are NOT the backlog: `/caddis:digress` parks an interrupted task on a
+stack (`/caddis:resume` pops it), and `.caddis/backlog/` is written by **docket** as a projection of
+its board — never hand-edit that one.
+
+### Write environment facts down in the same turn
+**If you learn a fact about the environment, write it to `.caddis/kb/environment-map.md` before you
+use it.** A hostname, which login actually works, where another team's repo really lives, what a
+folder actually contains, which port a service listens on.
+
+A fact that exists only in a conversation is gone when that session ends, and the next session
+cannot know it was ever said. One project paid for the same three facts repeatedly — ten minutes
+lost to one SQL login **three separate times**, and a session that searched the filesystem for a
+repository that lived in Gitea while the project's own docs cited a file inside it by path.
+
+`relay.md` is not the place: it is current state and gets rewritten at every handover, so stable
+facts drown in the churn. **Never record a secret** — record where the credential lives and which
+account works, never the value.
+
+### Doc discipline (the KB can't silently rot)
+Read `.caddis/kb/DOC-MAP.md` first to find the right reference doc, then read it on demand. When you
+add/rename/remove a route or a curated reference doc, update `UI_PAGE_GUIDE.md` / `DOC-MAP.md` in the
+**same change** — `scripts/check_doc_coverage.py` runs at the pre-push gate and **blocks** on a missing
+route or a dangling doc-map link.
+
+## Conventions & resources (plain files — read whatever the agent)
+- Per-area detail: `src/AGENTS.md`, `frontend/AGENTS.md`, `tests/AGENTS.md` (as present). Each folder's
+  `CLAUDE.md` is a shim that imports its `AGENTS.md`.
+- Reusable how-to: the caddis plugin's skills, invoked by name (e.g. `fastapi-dev`) — auto-triggered
+  when a task matches; not a project file. Codex/agy read the skill markdown directly.
+- Subagent briefs reusable as checklists: `.claude/agents/{tester,code-reviewer,preflight}.md`
+  (if vendored under `.claude/`, else provided by the caddis plugin).
+- Commands as runnable procedures: `.claude/commands/{feature-plan,tdd,prd,handoff}.md`
+  (if vendored under `.claude/`, else provided by the caddis plugin).
+- Commits: Conventional Commits. Terse output. Never commit secrets.

package/bundles/manifest.json

@@ -1,7 +1,8 @@
 {
-  "poolVersion": "1.3.75",
+  "poolVersion": "1.3.83",
   "bundles": {
-    "antigravity-plugin": "1.3.75",
-    "antigravity-plugin-extras": "1.3.24"
+    "antigravity-plugin": "1.3.83",
+    "antigravity-plugin-extras": "1.3.26",
+    "codex": "unknown"
   }
 }