redeploy 0.2.53__tar.gz → 0.2.55__tar.gz

{redeploy-0.2.53 → redeploy-0.2.55}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: redeploy
-Version: 0.2.53
+Version: 0.2.55
 Summary: Infrastructure migration toolkit: detect → plan → apply
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
@@ -32,17 +32,17 @@ Requires-Dist: mcp>=1.0; extra == "mcp"
 
 ## AI Cost Tracking
 
-![PyPI](https://img.shields.io/badge/pypi-costs-blue) ![Version](https://img.shields.io/badge/version-0.2.53-blue) ![Python](https://img.shields.io/badge/python-3.9+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
-![AI Cost](https://img.shields.io/badge/AI%20Cost-$7.50-orange) ![Human Time](https://img.shields.io/badge/Human%20Time-28.0h-blue) ![Model](https://img.shields.io/badge/Model-openrouter%2Fqwen%2Fqwen3--coder--next-lightgrey)
+![PyPI](https://img.shields.io/badge/pypi-costs-blue) ![Version](https://img.shields.io/badge/version-0.2.55-blue) ![Python](https://img.shields.io/badge/python-3.9+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
+![AI Cost](https://img.shields.io/badge/AI%20Cost-$7.50-orange) ![Human Time](https://img.shields.io/badge/Human%20Time-28.5h-blue) ![Model](https://img.shields.io/badge/Model-openrouter%2Fqwen%2Fqwen3--coder--next-lightgrey)
 
-- 🤖 **LLM usage:** $7.5000 (73 commits)
-- 👤 **Human dev:** ~$2802 (28.0h @ $100/h, 30min dedup)
+- 🤖 **LLM usage:** $7.5000 (75 commits)
+- 👤 **Human dev:** ~$2853 (28.5h @ $100/h, 30min dedup)
 
 Generated on 2026-04-21 using [openrouter/qwen/qwen3-coder-next](https://openrouter.ai/qwen/qwen3-coder-next)
 
 ---
 
-![PyPI](https://img.shields.io/badge/pypi-redeploy-blue) ![Version](https://img.shields.io/badge/version-0.2.53-blue) ![Python](https://img.shields.io/badge/python-3.10+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
+![PyPI](https://img.shields.io/badge/pypi-redeploy-blue) ![Version](https://img.shields.io/badge/version-0.2.55-blue) ![Python](https://img.shields.io/badge/python-3.10+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
 
 Infrastructure migration and device deploy toolkit — VPS, Raspberry Pi kiosk, Podman Quadlet, k3s.
 

{redeploy-0.2.53 → redeploy-0.2.55}/README.md

@@ -3,17 +3,17 @@
 
 ## AI Cost Tracking
 
-![PyPI](https://img.shields.io/badge/pypi-costs-blue) ![Version](https://img.shields.io/badge/version-0.2.53-blue) ![Python](https://img.shields.io/badge/python-3.9+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
-![AI Cost](https://img.shields.io/badge/AI%20Cost-$7.50-orange) ![Human Time](https://img.shields.io/badge/Human%20Time-28.0h-blue) ![Model](https://img.shields.io/badge/Model-openrouter%2Fqwen%2Fqwen3--coder--next-lightgrey)
+![PyPI](https://img.shields.io/badge/pypi-costs-blue) ![Version](https://img.shields.io/badge/version-0.2.55-blue) ![Python](https://img.shields.io/badge/python-3.9+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
+![AI Cost](https://img.shields.io/badge/AI%20Cost-$7.50-orange) ![Human Time](https://img.shields.io/badge/Human%20Time-28.5h-blue) ![Model](https://img.shields.io/badge/Model-openrouter%2Fqwen%2Fqwen3--coder--next-lightgrey)
 
-- 🤖 **LLM usage:** $7.5000 (73 commits)
-- 👤 **Human dev:** ~$2802 (28.0h @ $100/h, 30min dedup)
+- 🤖 **LLM usage:** $7.5000 (75 commits)
+- 👤 **Human dev:** ~$2853 (28.5h @ $100/h, 30min dedup)
 
 Generated on 2026-04-21 using [openrouter/qwen/qwen3-coder-next](https://openrouter.ai/qwen/qwen3-coder-next)
 
 ---
 
-![PyPI](https://img.shields.io/badge/pypi-redeploy-blue) ![Version](https://img.shields.io/badge/version-0.2.53-blue) ![Python](https://img.shields.io/badge/python-3.10+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
+![PyPI](https://img.shields.io/badge/pypi-redeploy-blue) ![Version](https://img.shields.io/badge/version-0.2.55-blue) ![Python](https://img.shields.io/badge/python-3.10+-blue) ![License](https://img.shields.io/badge/license-Apache--2.0-green)
 
 Infrastructure migration and device deploy toolkit — VPS, Raspberry Pi kiosk, Podman Quadlet, k3s.
 

{redeploy-0.2.53 → redeploy-0.2.55}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "redeploy"
-version = "0.2.53"
+version = "0.2.55"
 description = "Infrastructure migration toolkit: detect → plan → apply"
 readme = "README.md"
 requires-python = ">=3.11"

{redeploy-0.2.53 → redeploy-0.2.55}/redeploy/__init__.py

@@ -7,7 +7,7 @@ Public API (stable from 0.2.0, semver guaranteed)::
 
 Everything not listed in ``__all__`` is internal and may change without notice.
 """
-__version__ = "0.2.53"
+__version__ = "0.2.55"
 
 # ── Core models ───────────────────────────────────────────────────────────────
 from .models import (  # noqa: F401

redeploy-0.2.55/redeploy/heal/__init__.py

@@ -0,0 +1,37 @@
+"""redeploy.heal — Self-healing runner with LiteLLM auto-repair.
+
+This package has been refactored from the monolithic ``redeploy.heal`` module.
+For backward compatibility all public names remain importable from
+``redeploy.heal``.
+"""
+from __future__ import annotations
+
+from .loop_detector import HealAbort, HealLoopDetector
+from .log_writer import write_repair_log
+from .hint_provider import (
+    DIAG_COMMANDS,
+    KNOWN_CONSTRAINTS,
+    collect_diagnostics,
+    ask_llm,
+    apply_fix_to_spec,
+    parse_failed_step,
+)
+from .decider import Action, Decision, decide_after_failure, format_decision_message
+from .runner import HealRunner
+
+__all__ = [
+    "HealAbort",
+    "HealLoopDetector",
+    "write_repair_log",
+    "DIAG_COMMANDS",
+    "KNOWN_CONSTRAINTS",
+    "collect_diagnostics",
+    "ask_llm",
+    "apply_fix_to_spec",
+    "parse_failed_step",
+    "Action",
+    "Decision",
+    "decide_after_failure",
+    "format_decision_message",
+    "HealRunner",
+]

redeploy-0.2.55/redeploy/heal/decider.py

@@ -0,0 +1,85 @@
+"""Heal decision policy — retry / skip / abort logic."""
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass
+from enum import Enum
+from pathlib import Path
+from typing import Optional
+
+
+class Action(Enum):
+    RETRY = "retry"
+    SKIP = "skip"
+    ABORT = "abort"
+
+
+@dataclass(frozen=True)
+class Decision:
+    action: Action
+    reason: str = ""
+
+
+def decide_after_failure(
+    *,
+    attempt: int,
+    max_retries: int,
+    failed_step: Optional[str],
+    loop_detected: bool,
+    llm_error: bool,
+    spec_patched: bool,
+) -> Decision:
+    """Return the next action for the heal loop.
+
+    Parameters
+    ----------
+    attempt:
+        1-based attempt counter inside the heal loop.
+    max_retries:
+        Maximum retries configured by the user.
+    failed_step:
+        Step identifier, or *None* if the parser could not identify it.
+    loop_detected:
+        *True* when :class:`HealLoopDetector` flagged a repeating hint.
+    llm_error:
+        *True* when the LLM call itself failed (not a fixable step failure).
+    spec_patched:
+        *True* when the LLM produced a patch and it was applied.
+
+    Returns
+    -------
+    Decision
+    """
+    if loop_detected:
+        return Decision(
+            Action.ABORT,
+            f"Loop detected: same hint for {failed_step!r} ≥ 3 times in a row",
+        )
+    if failed_step is None:
+        return Decision(Action.ABORT, "Cannot identify failed step — giving up")
+    if llm_error:
+        return Decision(
+            Action.ABORT if attempt >= max_retries else Action.RETRY,
+            "LLM error — retrying may help if transient",
+        )
+    if attempt >= max_retries:
+        return Decision(
+            Action.ABORT,
+            f"Max retries ({max_retries}) exhausted for {failed_step!r}",
+        )
+    if spec_patched:
+        return Decision(Action.RETRY, "Spec patched — re-running migration")
+    return Decision(
+        Action.SKIP,
+        "LLM did not produce an applicable patch — skipping step",
+    )
+
+
+def format_decision_message(decision: Decision, step_id: str) -> str:
+    """Human-readable log / console message for a decision."""
+    emoji = {
+        Action.RETRY: "🔄",
+        Action.SKIP: "⏭️",
+        Action.ABORT: "🛑",
+    }
+    return f"{emoji[decision.action]} {decision.action.value.upper()}: {decision.reason}"

redeploy-0.2.55/redeploy/heal/hint_provider.py

@@ -0,0 +1,203 @@
+"""LLM hint generation and spec patching utilities."""
+from __future__ import annotations
+
+import os
+import re
+import textwrap
+from pathlib import Path
+from typing import Optional
+
+# ---------------------------------------------------------------------------
+# Diagnostics catalogue -- targeted SSH commands per failed step ID
+# ---------------------------------------------------------------------------
+DIAG_COMMANDS: dict[str, list[str]] = {
+    "restart-chromium": [
+        "pgrep -fa chromium | head -5 || echo NO_CHROMIUM",
+        "ls /run/user/1000/wayland-* 2>/dev/null || echo NO_WAYLAND",
+        "systemctl --user status kiosk-chromium.service --no-pager -n 10 2>&1 || true",
+        "cat ~/c2004/logs/kiosk.log 2>/dev/null | tail -20 || echo NO_LOG",
+    ],
+    "assert-screen-kiosk-url": [
+        "pgrep -fa chromium | head -3 || echo NO_CHROMIUM",
+        "cat ~/c2004/scripts/kiosk-launch.sh 2>/dev/null || echo NO_SCRIPT",
+        "cat ~/c2004/logs/kiosk.log 2>/dev/null | tail -30 || echo NO_LOG",
+    ],
+    "assert-screen-backend-healthy": [
+        "systemctl --user status c2004-backend.service --no-pager -n 20",
+        "journalctl --user -u c2004-backend.service --no-pager -n 20 2>/dev/null",
+        "curl -sv http://localhost:8000/api/v3/health 2>&1 | tail -10",
+    ],
+    "e2e-svg-icons": [
+        "podman exec c2004-frontend ls /usr/share/nginx/html/icons/ 2>/dev/null | head -10",
+        "curl -sI http://localhost:8100/icons/sprite.svg | head -8",
+        "podman exec c2004-frontend cat /etc/nginx/conf.d/default.conf 2>/dev/null | head -30",
+    ],
+    "e2e-api-endpoints": [
+        "systemctl --user status c2004-backend.service --no-pager -n 5",
+        "curl -sv http://localhost:8000/api/v3/health 2>&1 | tail -15",
+        "journalctl --user -u c2004-backend.service --no-pager -n 15 2>/dev/null",
+    ],
+    "e2e-kiosk-page-load": [
+        "curl -sf http://localhost:8100/ 2>/dev/null | head -20",
+        "podman exec c2004-frontend ls /usr/share/nginx/html/ | head -10",
+    ],
+    "_default": [
+        "systemctl --user list-units --type=service 2>/dev/null | grep -v '@' | tail -15",
+        "podman ps -a --format 'table {{.Names}}\\t{{.Status}}' 2>/dev/null",
+        "ss -tlnp 2>/dev/null | grep -E '8000|8100|8202' || echo PORTS_FREE",
+        "journalctl --user --no-pager -n 10 2>/dev/null",
+    ],
+}
+
+# Known constraints injected into every LLM prompt
+KNOWN_CONSTRAINTS = """
+Known constraints for this deployment target (Raspberry Pi 5, Podman Quadlet, labwc Wayland):
+- pkill/killall chromium via SSH = exit 255 (kills Wayland session and drops SSH)
+- Chromium requires: WAYLAND_DISPLAY=wayland-0 XDG_RUNTIME_DIR=/run/user/1000
+- Use systemd-run --user --collect to start GUI apps without blocking SSH
+- I2C device 0x45 on bus 11 is owned by kernel waveshare driver -- use || true
+- Redirecting /dev/null in heredoc via SSH causes exit 255
+- SQLite DBs at: /data/main/identification.db, /data/menu/menu.db, /data/scenario.db
+- alembic binary may not exist in container -- use python3 -c with sqlite3 module
+- Quadlet unit files go to ~/.config/containers/systemd/
+- Services: c2004-backend, c2004-frontend, c2004-firmware, c2004-reverse-proxy
+"""
+
+
+def _ssh(host: str, command: str) -> tuple[int, str]:
+    import subprocess
+    result = subprocess.run(
+        ["ssh", "-o", "ConnectTimeout=10", "-o", "BatchMode=yes", host, command],
+        capture_output=True, text=True, timeout=30,
+    )
+    return result.returncode, ((result.stdout or "") + (result.stderr or "")).strip()
+
+
+def collect_diagnostics(host: str, failed_step: str) -> str:
+    """Run targeted SSH diagnostics for a failed step, return combined output."""
+    cmds = DIAG_COMMANDS.get(failed_step, DIAG_COMMANDS["_default"])
+    parts = []
+    for cmd in cmds:
+        rc, out = _ssh(host, cmd)
+        parts.append(f"$ {cmd}\n{out}")
+    return "\n\n".join(parts)
+
+
+def ask_llm(
+    failed_step: str,
+    step_output: str,
+    diag: str,
+    spec_text: str,
+    fix_hint: str = "",
+) -> str:
+    """Ask LiteLLM to propose a fixed YAML block for the failed step."""
+    try:
+        import litellm
+    except ImportError:
+        return ""
+
+    model = os.getenv("LLM_MODEL", "openrouter/qwen/qwen3-coder-next")
+    api_key = os.getenv("OPENROUTER_API_KEY") or os.getenv("OPENAI_API_KEY")
+    api_base = "https://openrouter.ai/api/v1" if "openrouter" in model else None
+
+    hint_section = f"\n## User-reported issue:\n{fix_hint}\n" if fix_hint else ""
+
+    prompt = textwrap.dedent(f"""
+        You are an expert in Raspberry Pi 5 deployments with Podman Quadlet, labwc Wayland compositor and Chromium kiosk.
+
+        ## Failed step: `{failed_step}`
+        {hint_section}
+        ### Step output (error):
+        ```
+        {step_output[:2000]}
+        ```
+
+        ### SSH diagnostics collected after failure:
+        ```
+        {diag[:3000]}
+        ```
+
+        ### Current spec file (markpact YAML steps):
+        ```yaml
+        {spec_text[:5000]}
+        ```
+
+        {KNOWN_CONSTRAINTS}
+
+        ## Task:
+        Fix ONLY the step `{failed_step}`. Return ONLY the corrected YAML block:
+
+        ```yaml
+        - id: {failed_step}
+          action: ssh_cmd
+          description: "..."
+          command: |
+            <corrected command>
+        ```
+
+        Nothing else. No explanation outside the code block.
+    """).strip()
+
+    try:
+        kwargs: dict = dict(
+            model=model,
+            messages=[{"role": "user", "content": prompt}],
+            temperature=0.1,
+            max_tokens=1500,
+        )
+        if api_key:
+            kwargs["api_key"] = api_key
+        if api_base:
+            kwargs["api_base"] = api_base
+
+        resp = litellm.completion(**kwargs)
+        return resp.choices[0].message.content.strip()
+    except Exception as e:
+        return f"# LLM error: {e}"
+
+
+def apply_fix_to_spec(spec_path: Path, failed_step: str, llm_response: str) -> bool:
+    """Extract YAML block from LLM response and patch it into the spec file."""
+    m = re.search(r"```ya?ml\s*(.*?)```", llm_response, re.DOTALL)
+    if m:
+        new_block = m.group(1).strip()
+    else:
+        m = re.search(
+            rf"(- id: {re.escape(failed_step)}.+?)(?=\n  - id:|\Z)",
+            llm_response, re.DOTALL
+        )
+        if not m:
+            return False
+        new_block = m.group(1).strip()
+
+    text = spec_path.read_text()
+    pattern = rf"(  - id: {re.escape(failed_step)}\n(?:(?!  - id:).)*)"
+    match = re.search(pattern, text, re.DOTALL)
+    if not match:
+        return False
+
+    indented = "\n".join("  " + line if line else "" for line in new_block.splitlines())
+    spec_path.write_text(text[: match.start()] + indented + "\n" + text[match.end():])
+    return True
+
+
+def parse_failed_step(executor_summary: str, executor=None) -> tuple[Optional[str], str]:
+    """Extract (step_id, step_output) from executor state or summary string."""
+    if executor is not None:
+        state = getattr(executor, "state", None)
+        if state and state.failed_step_id:
+            results = getattr(executor, "_results", {})
+            step_out = results.get(state.failed_step_id, {})
+            if isinstance(step_out, dict):
+                out = step_out.get("output") or step_out.get("error") or ""
+            else:
+                out = str(step_out)
+            return state.failed_step_id, out
+
+    m = re.search(
+        r"Step failed: \[([^\]]+)\].*?exit=\d+:?\s*(.*?)(?=\n\d{{2}}:\d{{2}}|\Z)",
+        executor_summary, re.DOTALL
+    )
+    if m:
+        return m.group(1), m.group(2).strip()
+    return None, ""

redeploy-0.2.55/redeploy/heal/log_writer.py

@@ -0,0 +1,51 @@
+"""REPAIR_LOG.md writing utilities."""
+from __future__ import annotations
+
+import datetime
+from pathlib import Path
+
+
+_LOG_HEADER = "# Repair Log\n\nAuto-repairs by ``redeploy run --heal``.\n\n"
+
+
+def write_repair_log(
+    spec_path: Path,
+    version: str,
+    repairs: list[dict],
+) -> None:
+    """Append an entry to *REPAIR_LOG.md* adjacent to the spec file.
+
+    Parameters
+    ----------
+    spec_path:
+        Path to the spec that was healed.
+    version:
+        Current project version (written into the log heading).
+    repairs:
+        List of repair dicts produced by :class:`HealRunner`.
+    """
+    log_path = spec_path.parent / "REPAIR_LOG.md"
+    now = datetime.datetime.now().strftime("%Y-%m-%d %H:%M")
+    lines = [f"## {version} -- {now}\n"]
+    if repairs:
+        lines.append("### LLM repairs\n")
+        for r in repairs:
+            lines.append(f"- **`{r['step']}`**: {r.get('summary', 'fixed')}\n")
+            if r.get("diag_hint"):
+                lines.append(f"  - hint: `{r['diag_hint'][:120]}`\n")
+    else:
+        lines.append("Deployment completed without repairs.\n")
+    lines.append("\n")
+    entry = "".join(lines)
+
+    if log_path.exists():
+        existing = log_path.read_text()
+        if "# " in existing:
+            idx = existing.index("\n", existing.index("# ")) + 1
+            content = existing[:idx] + "\n" + entry + existing[idx:]
+        else:
+            content = entry + existing
+    else:
+        content = _LOG_HEADER + entry
+
+    log_path.write_text(content)

redeploy-0.2.55/redeploy/heal/loop_detector.py

@@ -0,0 +1,61 @@
+"""Heal loop detection — prevents non-converging auto-repair cycles."""
+from __future__ import annotations
+
+from collections import defaultdict
+from dataclasses import dataclass, field
+
+
+class HealAbort(Exception):
+    """Raised when a heal loop is detected and retries must stop."""
+
+    def __init__(self, step_id: str, hint: str, repeats: int) -> None:
+        self.step_id = step_id
+        self.hint = hint
+        self.repeats = repeats
+        super().__init__(
+            f"Heal loop detected for {step_id!r}: "
+            f"same hint repeated {repeats}x"
+        )
+
+
+@dataclass
+class HealLoopDetector:
+    """Detect repeated non-converging heal hints for a given step.
+
+    Parameters
+    ----------
+    max_identical_hints:
+        Number of consecutive identical hints that trigger a loop abort.
+        Default is 3 because after two identical retries the third
+        almost never brings new information.
+    """
+
+    max_identical_hints: int = 3
+    _history: dict[str, list[str]] = field(default_factory=lambda: defaultdict(list))
+
+    def observe(self, step_id: str, hint: str) -> bool:
+        """Return *True* when the latest hints indicate a heal loop.
+
+        A loop is defined as ``max_identical_hints`` identical non-empty
+        hints in a row for the same failed step.
+        """
+        normalized = (hint or "").strip()
+        if not normalized:
+            return False
+
+        history = self._history[step_id]
+        history.append(normalized)
+
+        if len(history) < self.max_identical_hints:
+            return False
+
+        recent = history[-self.max_identical_hints :]
+        return all(item == recent[0] for item in recent)
+
+    def reset(self, step_id: str) -> None:
+        """Clear history for *step_id* (e.g. after a successful retry)."""
+        self._history.pop(step_id, None)
+
+    def reset_all(self) -> None:
+        """Clear all history."""
+        self._history.clear()