task-harbor 0.7.0__tar.gz → 0.8.0__tar.gz

task_harbor-0.8.0/PKG-INFO

@@ -0,0 +1,134 @@
+Metadata-Version: 2.4
+Name: task-harbor
+Version: 0.8.0
+Summary: Claude and GPT backlog orchestrator for planning, implementation, QA, escalation, and periodic code review.
+Author: rhit-nayeems
+License-Expression: MIT
+Project-URL: Repository, https://github.com/rhit-nayeems/task-harbor
+Project-URL: Issues, https://github.com/rhit-nayeems/task-harbor/issues
+Requires-Python: >=3.11
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: textual<2,>=1
+Provides-Extra: dev
+Requires-Dist: pytest>=8; extra == "dev"
+Provides-Extra: ui
+Requires-Dist: textual<2,>=1; extra == "ui"
+Dynamic: license-file
+
+# Task Harbor
+
+Run a Markdown backlog with Claude Code or Codex. Task Harbor plans tasks,
+implements them in Git worktrees, runs your checks, and collects the results
+on a local branch for review. It does not push changes.
+
+![Terminal dashboard with sample run data](https://raw.githubusercontent.com/rhit-nayeems/task-harbor/main/docs/terminal-dashboard.svg)
+
+## Install
+
+Requires Python 3.11+ and Git. To run agent tasks, install and sign in to
+Claude Code or Codex.
+
+```sh
+python -m pip install task-harbor
+```
+
+Or use `pipx install task-harbor` for an isolated installation.
+If the command isn't on PATH, use `python -m task_harbor` with the same arguments.
+Upgrade with `python -m pip install --upgrade task-harbor` or `pipx upgrade task-harbor`.
+
+## Quick start
+
+Try a scripted bug fix first. No provider account or model calls are needed;
+the demo creates a separate sample repository and runs real local tests:
+
+```sh
+task-harbor demo
+```
+
+From a Git repository with at least one commit and a configured Git author:
+
+```sh
+task-harbor setup
+```
+
+The wizard asks for your provider, model, test command, and call limit. It
+requires verification and defaults to 20 calls per run. It creates
+`.manager/config.json` and a sample `BACKLOG.md`, keeping any existing backlog.
+Replace the sample with your tasks:
+
+```markdown
+## BUG-001 — Fix empty search results
+- **Priority:** critical_bug
+- **Status:** UNPLANNED
+- **Dependencies:** —
+- **Acceptance criteria:**
+  - An empty search shows a message instead of crashing
+  - Existing search tests pass
+- **Next action:** Plan the task
+```
+
+You can change these settings in `.manager/config.json`. For GPT, use
+`"provider": "codex"` and set both dollar-budget fields to `null`.
+See [configuration](https://github.com/rhit-nayeems/task-harbor/blob/main/docs/CONFIGURATION.md)
+for the fields and provider setup.
+
+For example, add these settings for a Python project:
+
+```json
+{
+  "verification_commands": ["python -m pytest"],
+  "max_agent_calls": 100,
+  "max_task_agent_calls": 20
+}
+```
+
+Check the setup, then start:
+
+```sh
+task-harbor check
+task-harbor manage
+```
+
+`check` reads local configuration without calling a model. `manage` starts agent
+work and prints a run ID. Commit your project changes before starting; use
+`task-harbor preflight --run-checks` to test your configured checks first.
+`init` remains available for manual setup; its call limits default to unlimited.
+
+Open the dashboard with `task-harbor`. It shows saved runs and task progress;
+F1 opens help. Closing the dashboard does not stop a run.
+
+## Review and resume
+
+Replace `RUN_ID` with an ID from `task-harbor runs`:
+
+```sh
+task-harbor runs
+task-harbor review --run-id RUN_ID --html review.html
+task-harbor stop --run-id RUN_ID
+task-harbor resume --run-id RUN_ID
+```
+
+The review report contains task evidence and verification results. Inspect the
+changes on `manager/integration/RUN_ID` before merging into your own branch.
+`stop` pauses after current calls finish; `resume` uses the saved task requirements.
+
+Run `task-harbor guide` for setup help or `task-harbor --help` for all commands.
+To work on another directory, put `--repo "path/to/project"` before the command.
+
+Agents execute local commands. Git worktrees are not a security sandbox.
+
+## Documentation
+
+- [Configuration and providers](https://github.com/rhit-nayeems/task-harbor/blob/main/docs/CONFIGURATION.md)
+- [Checks, task agreements, and review](https://github.com/rhit-nayeems/task-harbor/blob/main/docs/VERIFICATION.md)
+- [Compare with a direct agent](https://github.com/rhit-nayeems/task-harbor/blob/main/EVALUATION.md)
+- [Run a pilot](https://github.com/rhit-nayeems/task-harbor/blob/main/docs/PILOT.md)
+- [Architecture](https://github.com/rhit-nayeems/task-harbor/blob/main/ARCHITECTURE.md) · [Technical status](https://github.com/rhit-nayeems/task-harbor/blob/main/TECHNICAL_STATUS.md) · [Changelog](https://github.com/rhit-nayeems/task-harbor/blob/main/CHANGELOG.md)
+
+## Contributing
+
+Bug reports, fixes, and documentation improvements are welcome.
+See [CONTRIBUTING.md](https://github.com/rhit-nayeems/task-harbor/blob/main/CONTRIBUTING.md) for development setup and tests.
+
+[MIT License](https://github.com/rhit-nayeems/task-harbor/blob/main/LICENSE)

task_harbor-0.8.0/README.md

@@ -0,0 +1,116 @@
+# Task Harbor
+
+Run a Markdown backlog with Claude Code or Codex. Task Harbor plans tasks,
+implements them in Git worktrees, runs your checks, and collects the results
+on a local branch for review. It does not push changes.
+
+![Terminal dashboard with sample run data](https://raw.githubusercontent.com/rhit-nayeems/task-harbor/main/docs/terminal-dashboard.svg)
+
+## Install
+
+Requires Python 3.11+ and Git. To run agent tasks, install and sign in to
+Claude Code or Codex.
+
+```sh
+python -m pip install task-harbor
+```
+
+Or use `pipx install task-harbor` for an isolated installation.
+If the command isn't on PATH, use `python -m task_harbor` with the same arguments.
+Upgrade with `python -m pip install --upgrade task-harbor` or `pipx upgrade task-harbor`.
+
+## Quick start
+
+Try a scripted bug fix first. No provider account or model calls are needed;
+the demo creates a separate sample repository and runs real local tests:
+
+```sh
+task-harbor demo
+```
+
+From a Git repository with at least one commit and a configured Git author:
+
+```sh
+task-harbor setup
+```
+
+The wizard asks for your provider, model, test command, and call limit. It
+requires verification and defaults to 20 calls per run. It creates
+`.manager/config.json` and a sample `BACKLOG.md`, keeping any existing backlog.
+Replace the sample with your tasks:
+
+```markdown
+## BUG-001 — Fix empty search results
+- **Priority:** critical_bug
+- **Status:** UNPLANNED
+- **Dependencies:** —
+- **Acceptance criteria:**
+  - An empty search shows a message instead of crashing
+  - Existing search tests pass
+- **Next action:** Plan the task
+```
+
+You can change these settings in `.manager/config.json`. For GPT, use
+`"provider": "codex"` and set both dollar-budget fields to `null`.
+See [configuration](https://github.com/rhit-nayeems/task-harbor/blob/main/docs/CONFIGURATION.md)
+for the fields and provider setup.
+
+For example, add these settings for a Python project:
+
+```json
+{
+  "verification_commands": ["python -m pytest"],
+  "max_agent_calls": 100,
+  "max_task_agent_calls": 20
+}
+```
+
+Check the setup, then start:
+
+```sh
+task-harbor check
+task-harbor manage
+```
+
+`check` reads local configuration without calling a model. `manage` starts agent
+work and prints a run ID. Commit your project changes before starting; use
+`task-harbor preflight --run-checks` to test your configured checks first.
+`init` remains available for manual setup; its call limits default to unlimited.
+
+Open the dashboard with `task-harbor`. It shows saved runs and task progress;
+F1 opens help. Closing the dashboard does not stop a run.
+
+## Review and resume
+
+Replace `RUN_ID` with an ID from `task-harbor runs`:
+
+```sh
+task-harbor runs
+task-harbor review --run-id RUN_ID --html review.html
+task-harbor stop --run-id RUN_ID
+task-harbor resume --run-id RUN_ID
+```
+
+The review report contains task evidence and verification results. Inspect the
+changes on `manager/integration/RUN_ID` before merging into your own branch.
+`stop` pauses after current calls finish; `resume` uses the saved task requirements.
+
+Run `task-harbor guide` for setup help or `task-harbor --help` for all commands.
+To work on another directory, put `--repo "path/to/project"` before the command.
+
+Agents execute local commands. Git worktrees are not a security sandbox.
+
+## Documentation
+
+- [Configuration and providers](https://github.com/rhit-nayeems/task-harbor/blob/main/docs/CONFIGURATION.md)
+- [Checks, task agreements, and review](https://github.com/rhit-nayeems/task-harbor/blob/main/docs/VERIFICATION.md)
+- [Compare with a direct agent](https://github.com/rhit-nayeems/task-harbor/blob/main/EVALUATION.md)
+- [Run a pilot](https://github.com/rhit-nayeems/task-harbor/blob/main/docs/PILOT.md)
+- [Architecture](https://github.com/rhit-nayeems/task-harbor/blob/main/ARCHITECTURE.md) · [Technical status](https://github.com/rhit-nayeems/task-harbor/blob/main/TECHNICAL_STATUS.md) · [Changelog](https://github.com/rhit-nayeems/task-harbor/blob/main/CHANGELOG.md)
+
+## Contributing
+
+Bug reports, fixes, and documentation improvements are welcome.
+See [CONTRIBUTING.md](https://github.com/rhit-nayeems/task-harbor/blob/main/CONTRIBUTING.md) for development setup and tests.
+
+[MIT License](https://github.com/rhit-nayeems/task-harbor/blob/main/LICENSE)

{task_harbor-0.7.0 → task_harbor-0.8.0}/opus_manager/cli.py

@@ -27,6 +27,10 @@ def build_parser() -> argparse.ArgumentParser:
         description="Run a coding backlog with Claude or GPT. With no command, open the dashboard.",
         formatter_class=argparse.RawDescriptionHelpFormatter,
         epilog="""Quick start (inside your Git repository):
+  task-harbor demo           Try a scripted example without a provider
+  task-harbor setup          Guided setup: provider, checks, and call limit
+
+Manual setup:
   task-harbor init           Create config and sample backlog
   # Edit BACKLOG.md and .manager/config.json
   task-harbor validate       Check tasks without model calls
@@ -39,11 +43,18 @@ Dashboard keys: F1 help, Tab next pane, arrows select, r refresh, q quit.
 Python fallback: python -m task_harbor
 """,
     )
-    p.add_argument("--version", action="version", version="%(prog)s 0.7.0")
+    p.add_argument("--version", action="version", version="%(prog)s 0.8.0")
     p.add_argument("--repo", type=_repo, default=Path.cwd(), help="Target Git repository (default: current directory)")
     sub = p.add_subparsers(dest="command")
     p.set_defaults(command="tui", run_id=None)
     sub.add_parser("guide", help="Show setup, usage, and troubleshooting instructions")
+    demo = sub.add_parser("demo", help="Try a scripted bug fix with real tests; no provider or account needed")
+    demo.add_argument("--output", type=Path, help="New directory for the sample repository (default: a temporary directory)")
+    setup = sub.add_parser("setup", help="Configure a first run with required checks and call limits; no model calls")
+    setup.add_argument("--provider", choices=["claude", "codex"])
+    setup.add_argument("--model", help="Model identifier available in your provider account")
+    setup.add_argument("--check", action="append", help="Verification command; repeat for multiple checks")
+    setup.add_argument("--max-agent-calls", type=int, help="Run-wide call limit (20 for a new configuration)")
     tui = sub.add_parser("tui", help="Open the live terminal dashboard")
     tui.add_argument("--run-id", help="Select a saved run")
 
@@ -95,6 +106,8 @@ Python fallback: python -m task_harbor
     feedback.add_argument("--task", required=True)
     feedback.add_argument("--outcome", choices=["accepted", "rejected", "needs_changes"], required=True)
     feedback.add_argument("--minutes", type=float, required=True)
+    feedback.add_argument("--supervision-minutes", type=float, help="Active time preparing/guiding this task, excluding review and repair")
+    feedback.add_argument("--repair-minutes", type=float, help="Active time fixing this task, excluding review and supervision")
     feedback.add_argument("--note", default="")
     compare = sub.add_parser("compare", help="Prepare or run an opt-in direct-agent versus Task Harbor comparison")
     compare.add_argument("--manifest", type=Path, required=True)
@@ -205,6 +218,12 @@ def main(argv: list[str] | None = None) -> int:
             from .help import GUIDE
             print(GUIDE)
             return 0
+        if args.command == "demo":
+            from .onboarding import demo
+            return demo(args.output)
+        if args.command == "setup":
+            from .onboarding import setup
+            return setup(repo, args)
         if args.command == "init":
             return cmd_init(repo)
         if getattr(args, "run_id", None) is not None and (not args.run_id or any(c not in "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789_-" for c in args.run_id)):

{task_harbor-0.7.0 → task_harbor-0.8.0}/opus_manager/help.py

@@ -1,5 +1,10 @@
 GUIDE = """TASK HARBOR — GETTING STARTED
 
+Try task-harbor demo first: a separate sample repository, a scripted fix,
+real tests, and an offline review report. No provider account or model calls.
+For guided configuration in your own repository, run task-harbor setup.
+It requires checks and defaults to a 20-call run limit. Manual setup follows:
+
 1. Open a terminal in your Git repository (with at least one commit).
    task-harbor init
 2. Edit BACKLOG.md: give each task an ID and acceptance criteria.
@@ -41,6 +46,8 @@ Press E in the dashboard to inspect a task, P for suggested checks.
    task-harbor resume --run-id RUN_ID
 Record your review outcome separately from model QA:
    task-harbor feedback --run-id RUN_ID --task TASK_ID --outcome accepted --minutes 5
+Add --supervision-minutes and --repair-minutes to measure total active effort.
+Omitted values are unknown. Avoid counting the same minutes in multiple entries.
 Compare against a direct agent with an operator-written task manifest:
    task-harbor compare --manifest comparison.json --output prepared
 Preparation calls no models. Add --live with a new output folder to run both arms.

task_harbor-0.8.0/opus_manager/onboarding.py

@@ -0,0 +1,139 @@
+"""First-run configuration and an explicitly scripted, provider-free example."""
+from __future__ import annotations
+
+import json
+import sys
+import tempfile
+from dataclasses import asdict
+from pathlib import Path
+
+from .backlog import render_backlog, write_template
+from .config import Config
+from .evidence import change_evidence
+from .gitops import ensure_local_exclude, ensure_repo, git, head
+from .models import BacklogTask, ManagerState, TaskStatus
+from .offline import read_config
+from .preflight import suggestions
+from .reporting import review_summary, run_report, write_html
+from .state import atomic_write_json, save_state
+from .verification import run_checks
+
+
+def setup(repo: Path, args) -> int:
+    ensure_repo(repo)
+    cfg = read_config(repo)  # Reading an existing configuration must not migrate it yet.
+    path = repo / ".manager/config.json"
+    existing = json.loads(path.read_text(encoding="utf-8")) if path.exists() else {}
+    interactive = sys.stdin.isatty()
+
+    def choose(value, question, default=None):
+        if value is not None:
+            return value
+        if not interactive:
+            if default is not None:
+                return default
+            raise ValueError("Supply --provider, --model and --check, or run setup in an interactive terminal.")
+        answer = input(f"{question}" + (f" [{default}]" if default is not None else "") + ": ").strip()
+        return answer or default
+
+    provider = choose(args.provider, "Provider (claude/codex)", cfg.provider)
+    # Do not silently carry a Claude model into a Codex configuration.
+    model_default = cfg.model if provider == cfg.provider else None
+    model = choose(args.model, "Model available in your account", model_default)
+    commands = args.check or cfg.verification_commands
+    if not commands:
+        hints = suggestions(repo)
+        if hints:
+            print("Detected commands (not executed): " + ", ".join(hints))
+        commands = [choose(None, "Test/check command")]
+    if not all(isinstance(c, str) and c.strip() for c in commands):
+        raise ValueError("At least one nonempty verification command is required.")
+    limit = args.max_agent_calls
+    if limit is None:
+        limit = int(choose(None, "Maximum agent calls per run", cfg.max_agent_calls or 20))
+    values = asdict(cfg)
+    values.update(provider=provider, model=model, verification_commands=commands,
+                  require_verification=True, max_agent_calls=limit)
+    if provider != cfg.provider:
+        values.update(effort="medium", max_run_cost_usd=None, max_call_cost_usd=None)
+    configured = Config(**values)  # Validate everything before writing anything.
+    ensure_local_exclude(repo, ".manager")
+    if configured.manager_dir != ".manager":
+        ensure_local_exclude(repo, configured.manager_dir)
+    backlog = repo / configured.backlog_file
+    if not backlog.exists():
+        backlog.parent.mkdir(parents=True, exist_ok=True)
+        write_template(backlog)
+    atomic_write_json(path, {**existing, **asdict(configured)})
+    print(f"Saved {path}. Required checks: {len(commands)}. Run call limit: {limit}.")
+    print("Harbor's local files are excluded through Git's local info/exclude; .gitignore was not changed.")
+    print(f"1. Replace the sample task in {configured.backlog_file} with one small, observable change.")
+    print("2. Run task-harbor check and resolve errors. Authentication is not tested.")
+    print("3. Commit your project changes, then run task-harbor preflight --run-checks.")
+    print("4. Run task-harbor manage when ready to use your provider account.")
+    print("No model calls or verification commands were executed. Call limits are not token or dollar limits.")
+    return 0
+
+
+def demo(output: Path | None = None) -> int:
+    repo = output.expanduser().absolute() if output else Path(tempfile.mkdtemp(prefix="task-harbor-demo-"))
+    if output:
+        repo.mkdir(parents=True, exist_ok=False)  # Never overwrite an existing directory.
+    print(f"Sample repository: {repo}")
+    print("Scripted demonstration: no agents, authentication, or model usage. Tests and Git commits are real.")
+    git(repo, "init", "-b", "main")
+    git(repo, "config", "user.name", "Task Harbor demo")
+    git(repo, "config", "user.email", "demo@example.invalid")
+    (repo / ".gitignore").write_text(".manager/\n__pycache__/\nreview.html\n", encoding="utf-8")
+    source = repo / "search.py"
+    source.write_text("def search(query):\n    return query.split()[0]\n", encoding="utf-8")
+    (repo / "test_search.py").write_text(
+        "import unittest\nfrom search import search\n\nclass SearchTests(unittest.TestCase):\n"
+        "    def test_empty(self):\n        self.assertEqual(search(''), 'No results')\n"
+        "    def test_normal(self):\n        self.assertEqual(search('hello world'), 'hello')\n", encoding="utf-8")
+    task = BacklogTask("BUG-001", "Handle an empty search", acceptance_criteria=[
+        "Empty queries return 'No results'; normal queries still return their first word."])
+    render_backlog(repo / "BACKLOG.md", {task.task_id: task})
+    git(repo, "add", ".")
+    git(repo, "-c", "commit.gpgsign=false", "commit", "-m", "Sample search with an empty-query bug")
+    base = head(repo)
+    branch = "manager/integration/demo-001"
+    git(repo, "checkout", "-b", branch)
+    command = f'"{sys.executable}" -m unittest -v'
+    before = run_checks([command], repo, timeout=30, limit=12000)
+    print("1. Reproduced the bug: " + ("FAIL (expected)" if before[0]["exit_code"] else "unexpected PASS"))
+    source.write_text("def search(query):\n    words = query.split()\n    return words[0] if words else 'No results'\n", encoding="utf-8")
+    git(repo, "add", "search.py")
+    git(repo, "-c", "commit.gpgsign=false", "commit", "-m", "Handle empty queries")
+    checkpoint = head(repo)
+    checks = run_checks([command], repo, timeout=30, limit=12000)
+    if before[0]["exit_code"] == 0 or checks[0]["exit_code"] != 0:
+        raise RuntimeError("Demo checks did not produce the expected failure followed by a pass; inspect test_search.py.")
+    print("2. Applied the scripted fix; both tests pass at the recorded checkpoint.")
+    cfg = Config(verification_commands=[command], require_verification=True, max_agent_calls=20)
+    run_dir = repo / ".manager/runs/demo-001"
+    evidence_path = run_dir / "qa.json"
+    atomic_write_json(evidence_path, {
+        "checkpoint_commit": checkpoint, "verification_level": "deterministic_checks_only",
+        "summary": "The scripted fix handles empty input without changing normal search behavior.",
+        "harness_checks": checks, "acceptance": [{"criterion_id": "AC-1", "criterion": task.acceptance_criteria[0],
+            "status": "PASS", "evidence": "test_empty and test_normal passed in the local unittest run."}],
+        "reproduction": {"before": {"status": "FAIL", "checks": before}, "after": checks},
+        "changes": change_evidence(repo, base, checkpoint, cfg),
+        "risks": ["Scripted example only. No model QA ran. These two tests do not establish production correctness."]})
+    task.status = TaskStatus.COMPLETE
+    task.checkpoint_commit = checkpoint
+    task.last_qa_path = str(evidence_path)
+    task.next_action = "Inspect the diff and record your human review."
+    state = ManagerState("demo-001", str(repo), branch, str(repo), "main", base,
+                         tasks={task.task_id: task}, stop_reason="Scripted demo finished; human review pending.")
+    save_state(run_dir / "state.json", state)
+    cfg.save(repo / ".manager/config.json")
+    report = run_report(run_dir)
+    write_html(report, repo / "review.html")
+    print("3. Review the evidence:\n" + "\n".join(review_summary(report["tasks"][0], [])))
+    print(f'\nOpen {repo / "review.html"} in your browser, or explore the dashboard:')
+    print(f'task-harbor --repo "{repo}" tui --run-id demo-001')
+    print(f'Full checks and diff: task-harbor --repo "{repo}" review --run-id demo-001')
+    print(f'For your own project: task-harbor --repo "path/to/project" setup')
+    return 0

{task_harbor-0.7.0 → task_harbor-0.8.0}/opus_manager/operator_commands.py

@@ -74,9 +74,16 @@ def run_operator_command(args, repo, cfg):
         else:
             if not math.isfinite(args.minutes) or args.minutes < 0:
                 raise ValueError("Review minutes must be finite and nonnegative")
+            supervision = getattr(args, "supervision_minutes", None)
+            repair = getattr(args, "repair_minutes", None)
+            if any(value is not None and (not math.isfinite(value) or value < 0) for value in (supervision, repair)):
+                raise ValueError("Supervision and repair minutes must be finite and nonnegative")
             path = run_dir / "feedback.json"
             entries = json.loads(path.read_text(encoding="utf-8")) if path.exists() else []
             entries.append({"task": args.task, "outcome": args.outcome, "active_review_minutes": args.minutes,
+                            "active_supervision_minutes": supervision, "active_repair_minutes": repair,
+                            "active_engineer_minutes": args.minutes + supervision + repair
+                            if supervision is not None and repair is not None else None,
                             "note": args.note, "checkpoint": task.checkpoint_commit,
                             "time": datetime.now(timezone.utc).isoformat()})
             atomic_write_json(path, entries)

{task_harbor-0.7.0 → task_harbor-0.8.0}/opus_manager/reporting.py

@@ -18,6 +18,41 @@ def task_evidence(run_dir: Path, task) -> dict:
     return json.loads(path.read_text(encoding="utf-8"))
 
 
+def review_summary(task: dict, feedback: list) -> list[str]:
+    """Describe recorded evidence without treating model claims as independent proof."""
+    evidence = task["evidence"]
+    files = evidence.get("changes", {}).get("files", [])
+    checks = evidence.get("harness_checks", [])
+    passed = sum(c.get("exit_code") == 0 and not c.get("timed_out") for c in checks)
+    checkpoint = evidence.get("checkpoint_commit")
+    matching = bool(checkpoint and checkpoint == task.get("checkpoint"))
+    lines = ["Changed files: " + (", ".join(f["path"] for f in files) or "No diff evidence recorded."),
+             f"Checks: {passed} passed, {len(checks) - passed} failed/timed out, {len(checks)} recorded.",
+             "Checkpoint evidence: " + ("matches the task checkpoint." if matching else "missing or does not match; verify before accepting.")]
+    if evidence.get("summary"):
+        lines.append("Recorded QA summary: " + str(evidence["summary"]))
+    acceptance = evidence.get("acceptance", [])
+    for index, criterion in enumerate(task.get("acceptance_criteria", []), 1):
+        item = next((a for a in acceptance if a.get("criterion_id") == f"AC-{index}" or a.get("criterion") == criterion), {})
+        lines.append(f"AC-{index}: {criterion}; recorded {item.get('status', 'NOT VERIFIED')}"
+                     + (f"; evidence: {item['evidence']}" if item.get("evidence") else "; no supporting evidence recorded."))
+    warnings = [f"{f['path']}: {flag}" for f in files for flag in f.get("warnings", [])]
+    if not checks:
+        lines.append("Needs attention: no executed checks recorded.")
+    for warning in warnings:
+        lines.append("Needs attention: " + warning)
+    for risk in evidence.get("risks", []):
+        lines.append("Needs attention: " + str(risk))
+    reviews = [f for f in feedback if f.get("task") == task["id"] and f.get("checkpoint") == task.get("checkpoint") and task.get("checkpoint")]
+    lines.append("Human review: " + (reviews[-1]["outcome"] + " for this checkpoint." if reviews else "pending for this checkpoint."))
+    if reviews:
+        total = reviews[-1].get("active_engineer_minutes")
+        lines.append("Recorded engineer time (latest feedback): " + (f"{total:g} minutes." if total is not None
+                     else "incomplete; include supervision and repair minutes as well as review."))
+    lines.append("Next: inspect the diff and acceptance evidence, then record feedback. Passing checks do not prove correctness.")
+    return lines
+
+
 def review_text(report: dict, task_id: str | None = None) -> str:
     lines = [f'Task Harbor review: {report["run_id"]}', f'Branch: {report["integration_branch"]}']
     selected = [t for t in report["tasks"] if task_id is None or t["id"] == task_id]
@@ -26,6 +61,7 @@ def review_text(report: dict, task_id: str | None = None) -> str:
     for task in selected:
         evidence = task["evidence"]
         lines += ["", f'{task["id"]}: {task["title"]}', f'Status: {task["status"]}',
+                  *review_summary(task, report.get("feedback", [])),
                   f'Checkpoint: {evidence.get("checkpoint_commit", "not recorded")}',
                   f'Verification: {evidence.get("verification_level", "legacy/unknown")}',
                   "Task agreement: " + json.dumps(task["contract"], indent=2),
@@ -58,6 +94,7 @@ def run_report(run_dir: Path) -> dict:
                        "dependencies": t.dependencies, "blocked_reason": t.blocked_reason,
                        "next_action": t.next_action, "generation": t.generation,
                        "checkpoint": t.checkpoint_commit,
+                       "acceptance_criteria": t.acceptance_criteria,
                        "contract": t.metadata.get("contract", {}),
                        "evidence": task_evidence(run_dir, t),
                        "usage": summarize({"calls": [c for c in usage["calls"] if c.get("task") == t.task_id]})}
@@ -69,13 +106,17 @@ def write_html(report: dict, output: Path) -> None:
     rows = "".join("<tr>" + "".join(f"<td>{esc(t[k])}</td>" for k in
                    ("id", "title", "status", "blocked_reason", "next_action")) + "</tr>" for t in report["tasks"])
     details = "".join(f'<details><summary>{esc(t["id"])}: {esc(t["title"])}</summary><pre>{esc(review_text(report, t["id"]))}</pre></details>' for t in report["tasks"])
+    summaries = "".join(f'<section><h3>{esc(t["id"])}: {esc(t["title"])}</h3><ul>' +
+                        "".join(f"<li>{esc(line)}</li>" for line in review_summary(t, report.get("feedback", []))) +
+                        "</ul></section>" for t in report["tasks"])
     output.write_text(f'''<!doctype html><html lang="en"><meta charset="utf-8">
 <meta name="viewport" content="width=device-width, initial-scale=1"><title>Task Harbor · Run report</title>
 <style>body{{font:16px system-ui;margin:3rem;background:#101923;color:#e2eaf2}}h1{{color:#68debf}}
 table{{border-collapse:collapse;width:100%}}td,th{{padding:1rem;text-align:left;border-bottom:1px solid #35414d}}
-pre{{white-space:pre-wrap}}.scroll{{overflow:auto}}p{{line-height:1.7}}</style>
+pre{{white-space:pre-wrap;overflow-wrap:anywhere}}.scroll{{overflow:auto}}p,li{{line-height:1.7}}body{{max-width:1100px;margin:2rem auto;padding:0 1rem}}section{{border:1px solid #35414d;padding:1rem;margin:1rem 0}}li{{overflow-wrap:anywhere}}</style>
 <h1>Task Harbor</h1><p>Run {esc(report["run_id"])} · {report["completed"]}/{report["total"]} complete</p>
 <p>{esc(report["integration_branch"])}<br>{esc(report["stop_reason"])}</p>
+<h2>Review summary</h2>{summaries}
 <h2>Usage</h2><pre>{esc(json.dumps(report["usage"], indent=2))}</pre>
 <p>Costs are CLI estimates. Calls without reported cost are unknown, not free. Snapshot; regenerate to refresh.</p>
 <div class="scroll"><table><thead><tr><th>Task</th><th>Title</th><th>Status</th><th>Blocked</th><th>Next action</th></tr></thead>

{task_harbor-0.7.0 → task_harbor-0.8.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "task-harbor"
-version = "0.7.0"
+version = "0.8.0"
 description = "Claude and GPT backlog orchestrator for planning, implementation, QA, escalation, and periodic code review."
 requires-python = ">=3.11"
 authors = [{name = "rhit-nayeems"}]