src/skill_add_detector.py

#!/usr/bin/env python3
"""
skill_add_detector.py -- PostToolUse hook: detect when a new skill is written to a skill dir.

When Claude writes a SKILL.md file (via Write/Edit tools), this detects it and
registers the new skill in the wiki catalog and index automatically.

Called by PostToolUse hook:
    python skill_add_detector.py --tool <name> --input <json>

If the file is longer than the configured line threshold, the hook converts it
to a micro-skill pipeline under the wiki's converted/ directory immediately.

Two-tier slug validation model
-------------------------------
Tier 1 — lenient, for files already on disk (``wiki_utils.validate_skill_name``):
    Pattern: ``^[a-zA-Z0-9][a-zA-Z0-9_.-]{0,127}$``
    Allows uppercase, underscores, and dots so legacy installed skill names
    continue to work during re-ingestion and wiki-sync.

Tier 2 — strict, for user-supplied input (``validate_user_supplied_slug`` here):
    Pattern: ``^[a-z0-9][a-z0-9-]{0,63}$``
    Lowercase + hyphens only, 64-char max.  Used whenever an attacker could
    control the skill name (hook-detected writes, CLI --add, path extraction).
    Stricter than Tier 1 to minimise attack surface on newly-created entries.
"""

import json
import os
import re
import sys
from pathlib import Path

try:
    from ctx_config import cfg as _cfg
    CLAUDE_DIR = _cfg.claude_dir
    WIKI_DIR = _cfg.wiki_dir
    REGISTRY_PATH = _cfg.skill_registry
    CATALOG_PATH = _cfg.catalog
    LINE_THRESHOLD = _cfg.line_threshold
    SKILLS_DIR = _cfg.skills_dir
except ImportError:
    CLAUDE_DIR = Path(os.path.expanduser("~/.claude"))
    WIKI_DIR = CLAUDE_DIR / "skill-wiki"
    REGISTRY_PATH = CLAUDE_DIR / "skill-registry.json"
    CATALOG_PATH = WIKI_DIR / "catalog.md"
    LINE_THRESHOLD = 180
    SKILLS_DIR = CLAUDE_DIR / "skills"

# Skill directory names must be lowercase alnum + hyphen, bounded length.
# Stricter than the telemetry _SKILL_NAME_RE because directory names on
# disk should follow the canonical slug convention (no dots, no uppercase).
_SKILL_DIR_RE = re.compile(r"^[a-z0-9][a-z0-9-]{0,63}$")


def validate_user_supplied_slug(name: str) -> str:
    """Tier-2 (strict) validator for user-supplied skill slugs.

    Enforces the canonical slug contract for newly-created entries:
    lowercase alphanumeric + hyphens only, 64-char max.  Raises
    ``ValueError`` if the name does not match, preventing attacker-controlled
    path components from reaching downstream filesystem operations.

    For files already accepted on disk use
    ``wiki_utils.validate_skill_name`` (Tier-1, lenient).
    """
    if not isinstance(name, str) or not _SKILL_DIR_RE.match(name):
        raise ValueError(
            f"invalid skill name extracted from path: {name!r} "
            f"(must match {_SKILL_DIR_RE.pattern})"
        )
    return name


SKILL_TRIGGERS = {"Write", "Edit"}  # Tools that could create a skill file


def load_registry() -> list[str]:
    """Load registered skill directories from skill-registry.json."""
    if not REGISTRY_PATH.exists():
        return [str(CLAUDE_DIR / "skills"), str(CLAUDE_DIR / "agents")]
    try:
        return json.loads(REGISTRY_PATH.read_text())["skill_dirs"]
    except Exception:
        return []


def extract_written_path(tool_name: str, tool_input: dict) -> str | None:
    """Extract the file path being written from tool input."""
    if tool_name == "Write":
        return tool_input.get("file_path")
    elif tool_name == "Edit":
        return tool_input.get("file_path")
    return None


def is_in_skill_dir(file_path: str, skill_dirs: list[str]) -> bool:
    """Check if a file path is inside one of the registered skill directories."""
    p = Path(file_path).resolve()
    for d in skill_dirs:
        try:
            p.relative_to(Path(d).resolve())
            return True
        except ValueError:
            continue
    return False


def count_lines(file_path: str) -> int:
    """Count lines in a file safely."""
    try:
        return len(Path(file_path).read_text(encoding="utf-8", errors="replace").splitlines())
    except Exception:
        return 0


def _escape_md_cell(s: str) -> str:
    """Escape a string for safe embedding in a markdown table cell.

    Replaces pipe characters (which break table structure), collapses
    newlines to spaces, and escapes backticks to prevent nested code spans.
    """
    s = s.replace("\r\n", " ").replace("\n", " ").replace("\r", " ")
    s = s.replace("|", r"\|")
    s = s.replace("`", r"\`")
    return s


def register_skill_in_catalog(file_path: str, skill_name: str, lines: int) -> None:
    """Append a new skill entry to catalog.md."""
    if not CATALOG_PATH.exists():
        return
    content = CATALOG_PATH.read_text(encoding="utf-8")
    # Match against the structured table row (e.g. "| fastapi-pro |"), NOT a
    # naive substring — otherwise a skill named "react" would be skipped if
    # catalog rows mention "react-pro" or paths containing "react".
    if f"| {skill_name} |" in content:
        return
    over_flag = "⚠" if lines > LINE_THRESHOLD else ""
    safe_path = _escape_md_cell(file_path)
    entry = f"| {skill_name} | skill | {lines} | {over_flag} | `{safe_path}` |"
    # Insert before the last line
    lines_list = content.splitlines()
    lines_list.append(entry)
    CATALOG_PATH.write_text("\n".join(lines_list) + "\n", encoding="utf-8")


def maybe_convert_to_micro_skill(
    skill_path: Path,
    skill_name: str,
    line_count: int,
) -> tuple[bool, str]:
    """Convert long hook-detected skills to the micro-skill pipeline."""
    if line_count <= LINE_THRESHOLD:
        return False, "below-threshold"
    try:
        from batch_convert import convert_skill

        output_dir = WIKI_DIR / "converted" / skill_name
        output_dir.mkdir(parents=True, exist_ok=True)
        result = convert_skill(skill_path, output_dir=output_dir)
    except Exception as exc:  # noqa: BLE001 - hook output should report failure, not crash Claude.
        return False, f"conversion failed: {exc}"
    if result.get("status") == "converted":
        return True, str(output_dir)
    return False, f"conversion skipped: {result.get('reason') or result.get('status')}"


def _parse_stdin_payload() -> tuple[str, dict]:
    """Read the PostToolUse JSON payload from stdin.

    Claude Code delivers the payload on stdin when --from-stdin is used,
    avoiding $CLAUDE_TOOL_INPUT interpolation into argv (shell injection risk).
    Returns (tool_name, tool_input).
    """
    try:
        raw = sys.stdin.read()
        if not raw.strip():
            return "unknown", {}
        data = json.loads(raw)
        if not isinstance(data, dict):
            return "unknown", {}
        tool_name = str(data.get("tool_name") or "unknown")
        tool_input = data.get("tool_input") or {}
        if not isinstance(tool_input, dict):
            tool_input = {}
        return tool_name, tool_input
    except (json.JSONDecodeError, OSError):
        return "unknown", {}


def main() -> None:
    import argparse
    parser = argparse.ArgumentParser()
    parser.add_argument("--tool", default="unknown")
    parser.add_argument("--input", default="{}")
    parser.add_argument(
        "--from-stdin",
        action="store_true",
        help="Read tool_name and tool_input from the JSON payload on stdin "
             "(safe alternative to --tool/--input that avoids shell injection)",
    )
    args = parser.parse_args()

    if args.from_stdin:
        tool_name, tool_input = _parse_stdin_payload()
    else:
        tool_name = args.tool
        try:
            tool_input = json.loads(args.input)
        except json.JSONDecodeError:
            sys.exit(0)

    if tool_name not in SKILL_TRIGGERS:
        sys.exit(0)

    file_path = extract_written_path(tool_name, tool_input)
    if not file_path:
        sys.exit(0)

    # Only care about SKILL.md files
    if not file_path.endswith("SKILL.md"):
        sys.exit(0)

    skill_dirs = load_registry()
    if not is_in_skill_dir(file_path, skill_dirs):
        sys.exit(0)

    # New skill detected in a skill directory.
    # Resolve to an absolute path before extracting the parent name so that
    # traversal sequences like "/tmp/../../etc/foo" are normalised away first.
    # We already know the resolved path is inside a registered skill dir
    # (is_in_skill_dir uses resolve() internally), so extracting .parent.name
    # from the resolved path gives us the real directory name.
    resolved_path = Path(file_path).resolve()
    raw_name = resolved_path.parent.name
    try:
        skill_name = validate_user_supplied_slug(raw_name)
    except ValueError:
        sys.exit(0)
    lines = count_lines(file_path)

    # Register in catalog
    register_skill_in_catalog(file_path, skill_name, lines)

    if lines > LINE_THRESHOLD:
        converted, detail = maybe_convert_to_micro_skill(
            resolved_path,
            skill_name,
            lines,
        )
        status = "converted" if converted else "conversion not completed"
        print(
            f"\n[skill-system] New skill '{skill_name}' has {lines} lines "
            f"(>{LINE_THRESHOLD}); micro-skill gate {status}: {detail}\n"
        )


if __name__ == "__main__":
    main()