FabledScribe/scripts/check_plugin.py

#!/usr/bin/env python3
"""Guards for `plugin/` — the one part of this repo that ships straight to users.

WHY THIS EXISTS. `plugin/` is not built into the Docker image. Installs fetch it
from this git repo via `.claude-plugin/marketplace.json`, so a push IS the
release for plugin content: no build, no gate, immediately fetchable. Two
separate defects have reached a live install through that path:

  - #2198 — all four hook scripts were inert (wrong env-var case, line-oriented
    `jq -rR`, line-oriented `cut -c`). No CI ran, because `plugin/**` wasn't in
    the workflow's `paths:` filter at all.
  - #2209 — the fix for #2198 shipped to `main` and still couldn't reach an
    install, because `plugin.json`'s version wasn't bumped and the installer
    compares versions to decide whether to refresh its cache.

The rule for the second one was already written down and was still missed. A
written rule that depends on being remembered during a long session is not a
control; this is.

shellcheck and jq are NOT in `ci-python` (verified against CI-runner's Dockerfile
and scripts/install-common.sh, not from memory — rule #37). CI installs both
per-job, which is what CI-runner's own docs/process.md prescribes for a dep with
a single consumer: "If only one project needs the dep, prefer that project
installing it per-job in their workflow — at least until a second consumer
arrives." Promotion into the image is filed as an issue there rather than
assumed here.

Both are optional at runtime: without shellcheck the lint step is SKIPPED and
says so, and without jq the smoke test is skipped. A skipped check announces
itself loudly, because a check that quietly no-ops is the failure mode this
whole file exists to prevent.

Usage:
    python3 scripts/check_plugin.py            # all checks
    python3 scripts/check_plugin.py --no-version   # skip the bump check
"""
from __future__ import annotations

import argparse
import json
import os
import re
import shutil
import subprocess
import sys
from pathlib import Path

ROOT = Path(__file__).resolve().parents[1]
PLUGIN_DIR = ROOT / "plugin"
HOOKS_DIR = PLUGIN_DIR / "hooks"
MANIFEST = PLUGIN_DIR / ".claude-plugin" / "plugin.json"

# ── What ships, and what decides what it says about itself ─────────────────
#
# ONE definition (#3127 §3, milestone 334 step 2). It has TWO consumers that
# need different granularities, and conflating them is the bug:
#
#   the workflow's `paths:` trigger   whole paths   should CI run at all?
#   the version check                 paths MINUS   should the version
#                                     the manifest  have moved?
#                                     `version`
#
# The second one is why this is not just a tuple of paths. `plugin.json` lives
# INSIDE `plugin/`, so a version bump is itself a change to the shipped set —
# and a check that reads the set naively then treats the bump as its own
# justification. Any bump passes, no bump fails, and it has proved nothing.
# `shipped_content_changed` below is the exclusion-aware reader.
#
# The exclusion is that ONE FIELD, never the whole file: `plugin.json` also
# carries description, mcpServers and userConfig, all of which reach an
# install and all of which matter. Excluding the file wholesale would mean a
# userConfig-only edit computes an unchanged version and never refreshes —
# #2209 again with a narrower trigger.
SHIPPED_PATHS = ("plugin", ".claude-plugin")

# Files that decide what a published artifact SAYS ABOUT ITSELF — kept as a
# table so the next artifact is a one-line addition rather than a third
# bespoke guard (#3127 §3). The membership test is NOT "is this copied into
# the artifact?" but "can changing this file change the published bytes, or
# what the artifact says about itself?" — FC learned that twice in four days
# (#3156, #3202), and a deriver is never in the COPY list.
#
# Note what is absent: a CHECKER does not belong here. Whatever validates a
# version decides whether the lane goes red, not what any artifact reports,
# so `check_plugin.py` itself is not a deriver — the plugin's mint script
# (milestone 334 step 3) will be, and adds its own row.
DERIVERS: dict[str, tuple[str, ...]] = {
    # The "Generate image tags and version" step computes the server image's
    # name, ordering key and channel (#3298).
    ".forgejo/workflows/ci.yml": ("server-image",),
}

failures: list[str] = []


def fail(msg: str) -> None:
    failures.append(msg)
    print(f"FAIL  {msg}")


def ok(msg: str) -> None:
    print(f"ok    {msg}")


def skip(msg: str) -> None:
    # Loud on purpose. A check that quietly does nothing is indistinguishable
    # from a check that passed — the exact confusion that let #2198 survive.
    print(f"SKIP  {msg}")


def hook_scripts() -> list[Path]:
    return sorted(HOOKS_DIR.glob("*.sh"))


# --- syntax ----------------------------------------------------------------

def check_syntax() -> None:
    """`bash -n` every hook. Catches nothing subtle, costs nothing, and a syntax
    error here means a hook that silently never runs."""
    for script in hook_scripts():
        proc = subprocess.run(
            ["bash", "-n", str(script)], capture_output=True, text=True
        )
        if proc.returncode != 0:
            fail(f"{script.relative_to(ROOT)}: bash -n — {proc.stderr.strip()}")
        else:
            ok(f"{script.relative_to(ROOT)}: syntax")


# --- known-bad patterns ----------------------------------------------------

# Each entry: (compiled pattern, short label, why it's wrong).
# These are the exact classes from #2198. They are deliberately specific — a
# broad shell linter belongs in the image, not hand-rolled here.
PATTERNS: list[tuple[re.Pattern, str, str]] = [
    (
        re.compile(r"CLAUDE_PLUGIN_OPTION_[a-z]"),
        "lowercase userConfig env var",
        "Claude Code exports userConfig to hooks as CLAUDE_PLUGIN_OPTION_<KEY> "
        "with the key UPPERCASED. The lowercase spelling reads as empty and the "
        "hook then does nothing, silently.",
    ),
    (
        # -R without -s: reads input line by line, so a multi-line payload is
        # encoded per line and joined with raw newlines. The class is a-r + t-z
        # (i.e. every letter EXCEPT `s`) so `-rR` is caught and `-sRr` is not —
        # an earlier a-q spelling silently excluded `r` and missed the real
        # defect, which is exactly the flag combination that shipped.
        re.compile(r"jq\s+-(?:[a-rt-zA-Z]*R[a-rt-zA-Z]*)\s"),
        "line-oriented jq -R",
        "jq -R reads input LINE BY LINE. Encoding a multi-line payload that way "
        "produces separate encoded lines joined by raw newlines — an invalid "
        "URL. Use -s (slurp) as well, e.g. `jq -sRr '@uri'`.",
    ),
    (
        re.compile(r"\|\s*cut\s+-c"),
        "line-oriented cut for a payload cap",
        "cut -c truncates EACH LINE, so it does not cap total size. Use "
        "`head -c N` to bound a payload.",
    ),
]


def check_patterns() -> None:
    for script in hook_scripts():
        text = script.read_text(encoding="utf-8", errors="replace")
        rel = script.relative_to(ROOT)
        hits = 0
        for line_no, line in enumerate(text.splitlines(), 1):
            # A line that only *documents* the trap is fine — several hooks now
            # carry a comment naming the wrong form so the next reader knows.
            if line.lstrip().startswith("#"):
                continue
            for pattern, label, why in PATTERNS:
                if pattern.search(line):
                    hits += 1
                    fail(f"{rel}:{line_no}: {label}\n        {line.strip()}\n        {why}")
        if not hits:
            ok(f"{rel}: no known-bad patterns")


# --- the version bump ------------------------------------------------------

# --- shellcheck ------------------------------------------------------------

def check_shellcheck() -> None:
    """Real shell linting, where available.

    The hand-rolled patterns above only know the bugs that already happened.
    This is what catches the next one.
    """
    exe = shutil.which("shellcheck")
    if not exe:
        skip("shellcheck not installed — install it to lint the hooks properly")
        return
    for script in hook_scripts():
        proc = subprocess.run(
            # -x FOLLOWS `# shellcheck source=` directives into the sourced
            # file. Without it the shared helpers in scribe_defs.sh are
            # invisible, so every variable they set reads as unassigned
            # (SC2154) and every bug inside them goes unlinted at the call
            # site — which is the opposite of what sharing them was for.
            [exe, "--severity=warning", "--shell=bash", "-x", str(script)],
            capture_output=True, text=True,
        )
        rel = script.relative_to(ROOT)
        if proc.returncode != 0:
            fail(f"{rel}: shellcheck\n{proc.stdout.strip()}")
        else:
            ok(f"{rel}: shellcheck")


# --- the fail-open contract ------------------------------------------------

# Every hook promises never to break the operator's session: unconfigured or
# unreachable, it exits 0. Unconfigured, the enrichment hooks are SILENT — no
# call was owed. scribe_session_context.sh is the exception by design — it
# always emits a static behavioural floor that needs no credentials and no
# network, so "silent" would be the wrong assertion for it.
#
# UNREACHABLE is different for the two write-path hooks since #2932: a call
# that was owed and did not come back is SAID, once per outage ("> Scribe did
# not answer …"), so a session can tell "checked, nothing there" from "never
# checked". That line — or silence, when the once-per-outage marker in
# ${TMPDIR:-/tmp}/scribe-priorart/ was set by a run in the last ten minutes —
# is the only output allowed with no working instance; anything else is a hook
# speaking on data it cannot have.
#
# This is the contract that made #2198 invisible for weeks, so it is worth
# pinning: the bug and the healthy no-results case looked identical from
# outside. #2932 is what finally makes the failure visible at the write; this
# check makes sure the fail-open behaviour stays deliberate rather than
# accidental.
# A symbol that exists nowhere, ASSEMBLED rather than written literally.
# The prior-art hook's local arm (#2280) fires with no credentials, so the
# silence assertion below needs a name the repo genuinely lacks. Two traps,
# both hit while writing this:
#   - `def f` matched real code, so the hook spoke and "silent" was asserting
#     the wrong thing;
#   - spelling the replacement out in full put `def <name>(` INTO this file,
#     so the smoke event defined the very symbol it claimed was absent.
# Concatenating keeps the contiguous string out of the source.
_ABSENT_SYM = "zz" + "_absent_" + "9f3a2b"

SMOKE_EVENTS: dict[str, str] = {
    "scribe_autoinject.sh": json.dumps(
        {"session_id": "smoke", "cwd": ".", "prompt": "a multi-line\nprompt\nhere"}
    ),
    "scribe_prior_art.sh": json.dumps(
        {"session_id": "smoke", "cwd": ".", "tool_name": "Edit",
         "tool_input": {"file_path": "src/x.py",
                        "new_string": f"def {_ABSENT_SYM}():\n    pass\n"}}
    ),
    "scribe_sync_processes.sh": json.dumps({"source": "startup"}),
    "scribe_session_context.sh": json.dumps({"source": "startup"}),
    # The after-write hook (#2901) diffs the working tree; on CI's clean
    # checkout there is nothing to report, so silence is the right assertion.
    # (On a dirty local tree with a definition just written it may speak —
    # that is the hook working, not a failure of the contract.)
    "scribe_after_write.sh": json.dumps(
        {"session_id": "smoke", "cwd": ".", "tool_name": "Bash",
         "tool_input": {"command": "true"}, "tool_response": {}}
    ),
    # The shared library is sourced, never run; executed bare it defines
    # functions and exits — silent by construction.
    "scribe_defs.sh": "",
}

# The one hook that legitimately produces output with no credentials.
STATIC_FLOOR = "scribe_session_context.sh"
# The hooks that say so when a configured instance does not answer (#2932).
OUTAGE_SPEAKERS = {"scribe_prior_art.sh", "scribe_after_write.sh"}
OUTAGE_LINE = "> Scribe did not answer the prior-art check"


def _run_hook(script: Path, event: str, env_extra: dict[str, str]) -> subprocess.CompletedProcess:
    env = {k: v for k, v in os.environ.items()
           if not k.startswith(("SCRIBE_", "CLAUDE_PLUGIN_OPTION_"))}
    env.update(env_extra)
    return subprocess.run(
        ["bash", str(script)], input=event, capture_output=True, text=True,
        env=env, timeout=30,
    )


def check_fail_open() -> None:
    if not shutil.which("jq"):
        # Without jq every hook bails at its first line, so this would pass
        # while exercising nothing. Say so rather than bank a green tick.
        skip("jq not installed — the hooks would exit at line 1, so this "
             "check would pass without testing anything")
        return

    scenarios = [
        ("unconfigured", {}),
        # Connection refused immediately — exercises the unreachable-instance
        # path without waiting on a real network timeout.
        ("unreachable", {"SCRIBE_URL": "http://127.0.0.1:1", "SCRIBE_TOKEN": "x"}),
    ]
    for script in hook_scripts():
        rel = script.relative_to(ROOT)
        event = SMOKE_EVENTS.get(script.name)
        if event is None:
            skip(f"{rel}: no smoke event defined")
            continue
        for label, env_extra in scenarios:
            try:
                proc = _run_hook(script, event, env_extra)
            except subprocess.TimeoutExpired:
                fail(f"{rel} [{label}]: hung — a hook must never block a session")
                continue
            if proc.returncode != 0:
                fail(f"{rel} [{label}]: exited {proc.returncode}, must be 0 — "
                     f"a recall aid may never fail the operator's action")
                continue
            out = proc.stdout.strip()
            if script.name == STATIC_FLOOR:
                # Emits its bundled static tier regardless; that floor is the
                # whole point of the two-tier design.
                if not out:
                    fail(f"{rel} [{label}]: emitted nothing — the static "
                         f"behavioural floor must survive having no credentials")
                else:
                    ok(f"{rel} [{label}]: exit 0, static floor present")
            elif out and label == "unreachable" and script.name in OUTAGE_SPEAKERS:
                # The only thing allowed here is the outage line itself.
                try:
                    ctx = json.loads(out)["hookSpecificOutput"]["additionalContext"]
                except (ValueError, KeyError, TypeError):
                    ctx = ""
                if ctx.startswith(OUTAGE_LINE):
                    ok(f"{rel} [{label}]: exit 0, says the instance did not answer")
                else:
                    fail(f"{rel} [{label}]: emitted output with no working instance "
                         f"that is not the outage line:\n        {out[:200]}")
            elif out:
                fail(f"{rel} [{label}]: emitted output with no working instance:\n"
                     f"        {out[:200]}")
            else:
                ok(f"{rel} [{label}]: exit 0, silent")


def check_local_prior_art_needs_no_instance() -> None:
    """The prior-art hook's local arm must answer with no credentials (#2280).

    The other arms ask Scribe what was RECORDED. This one asks the repo what
    EXISTS, which needs no instance — and that is the whole reason it catches
    the case the recorded arms structurally cannot: a helper nobody thought to
    record. If it ever silently starts depending on configuration, it stops
    covering that case and nothing else would notice.

    Paired with the silence assertion in check_fail_open, which uses a symbol
    that cannot exist. Together they pin both halves: silent when there is
    nothing to say, and speaking when there is — both with no instance at all.
    """
    script = HOOKS_DIR / "scribe_prior_art.sh"
    if not script.is_file() or not shutil.which("jq"):
        skip("prior-art local arm: hook or jq missing")
        return

    # A definition this repo really does contain, written into a DIFFERENT file
    # so the self-match exclusion doesn't suppress it.
    event = json.dumps({
        "session_id": "smoke", "cwd": ".", "tool_name": "Write",
        "tool_input": {
            "file_path": "scripts/_probe_not_real.py",
            "content": "def check_local_prior_art_needs_no_instance():\n    pass\n",
        },
    })
    try:
        proc = _run_hook(script, event, {})       # NO credentials, on purpose
    except subprocess.TimeoutExpired:
        fail("prior-art local arm: hung")
        return
    if proc.returncode != 0:
        fail(f"prior-art local arm: exited {proc.returncode}, must be 0")
    elif "already defined" not in proc.stdout:
        fail("prior-art local arm: found nothing for a symbol this repo "
             "defines, with no credentials — the arm that needs no instance "
             "has stopped working, and the recorded arms cannot cover for it")
    else:
        ok("prior-art local arm: answers with no instance configured")


def check_session_context_reports_its_version() -> None:
    """The SessionStart context must name the plugin version it is running.

    An install has two halves and only one self-updates: the marketplace clone
    pulls on its own, while the CACHE is what executes and refreshes only when
    the manifest version changes. So a shipped fix can sit unreached while
    inspecting the clone shows it present — the obvious debugging move
    misleads, and twice the only detector was a human saying "I don't think it
    updated" (#2209, #2220).

    Asserted WITHOUT credentials on purpose. The state most needing diagnosis
    is the one where the token never arrives, and a marker that vanished there
    would be missing exactly when it is wanted.
    """
    script = HOOKS_DIR / "scribe_session_context.sh"
    if not script.is_file() or not shutil.which("jq"):
        skip("version marker: hook or jq missing")
        return

    manifest_v = manifest_version()
    if manifest_v is None:
        fail("version marker: could not read the manifest version")
        return

    try:
        proc = _run_hook(script, json.dumps({"source": "startup"}), {})
    except subprocess.TimeoutExpired:
        fail("version marker: hook hung")
        return
    if proc.returncode != 0:
        fail(f"version marker: hook exited {proc.returncode}")
    elif manifest_v not in proc.stdout:
        fail(f"version marker: session context never names v{manifest_v} — "
             f"a stale install would be undetectable from the transcript")
    else:
        ok(f"version marker: session context reports v{manifest_v}, no credentials needed")


def _git(*args: str) -> tuple[int, str]:
    proc = subprocess.run(
        ["git", *args], capture_output=True, text=True, cwd=ROOT
    )
    return proc.returncode, (proc.stdout or proc.stderr).strip()


def manifest_text(ref: str | None = None) -> str | None:
    """The manifest's RAW TEXT at `ref`, or in the working tree when ref is None.

    Split out from `manifest_version` because the exclusion below needs every
    field except one, not the one field.
    """
    if ref is None:
        try:
            return MANIFEST.read_text()
        except OSError:
            return None
    rel = MANIFEST.relative_to(ROOT).as_posix()
    code, out = _git("show", f"{ref}:{rel}")
    return out if code == 0 else None


def manifest_version(ref: str | None = None) -> str | None:
    """The manifest version at `ref`, or in the working tree when ref is None."""
    text = manifest_text(ref)
    if text is None:
        return None
    try:
        return json.loads(text).get("version")
    except Exception:
        return None


# Distinct from None, which is a legitimate "this manifest does not exist".
_UNREADABLE = object()


def manifest_differs_beyond_version(a: str | None, b: str | None) -> bool:
    """Do two `plugin.json` texts differ in anything OTHER than `version`?

    THE exclusion, and it is kept pure — no git, no filesystem — because this
    is the half worth testing hard and it needs no repository to exercise.

    Compares PARSED objects rather than text, so reformatting, key reordering
    and whitespace do not read as content changes. `version` is dropped from
    both sides; everything else counts, which is what keeps a userConfig-only
    or mcpServers-only edit demanding a new version.

    Unreadable input answers True. The conservative direction is "demand a new
    version": a spurious bump costs one cache refresh, while a missed one is
    #2209 — the fix reaches the repo and stops there.
    """
    def without_version(text: str | None):
        if text is None:
            return None
        try:
            data = json.loads(text)
        except Exception:
            return _UNREADABLE
        if not isinstance(data, dict):
            return _UNREADABLE
        return {k: v for k, v in data.items() if k != "version"}

    left, right = without_version(a), without_version(b)
    if left is _UNREADABLE or right is _UNREADABLE:
        return True
    return left != right


def shipped_content_changed(base: str) -> tuple[bool | None, list[str]]:
    """Has anything that REACHES AN INSTALL changed against `base`?

    Returns `(changed, paths)`. `changed` is **None** when the question could
    not be answered — a caller must never read that as "no", which is the
    distinction #2663 cost weeks of zeroed telemetry to learn.

    The manifest is special-cased, not excluded: if it is the ONLY thing that
    moved and the only difference is `version`, nothing that reaches an
    install has changed. Any other manifest field, or any other file, counts.
    """
    code, out = _git("diff", "--name-only", base, "--", *SHIPPED_PATHS)
    if code != 0:
        return None, []
    paths = [p for p in out.splitlines() if p.strip()]
    if not paths:
        return False, []

    rel_manifest = MANIFEST.relative_to(ROOT).as_posix()
    if paths == [rel_manifest]:
        return manifest_differs_beyond_version(
            manifest_text(), manifest_text(base)
        ), paths
    return True, paths


def check_version_bump(base: str = "origin/main") -> None:
    """If shipped plugin content differs from `base`, the version must too.

    Stated against the BASE BRANCH rather than the last commit on purpose. A
    per-commit rule would demand a bump from every commit in a batch; what
    actually matters is that whatever reaches an install carries a version the
    installer can tell apart from the one already cached. One bump per batch,
    which is also how a human would do it.

    Reads the set through `shipped_content_changed`, so a commit whose ONLY
    change is the version field does not count as content moving. Without that
    the check is circular — the bump edits a file inside `plugin/`, which then
    reads as the change that justifies the bump.
    """
    code, _ = _git("rev-parse", "--verify", base)
    if code != 0:
        # Do NOT pass silently — a check that quietly no-ops is how this class
        # of bug survives in the first place.
        fail(
            f"cannot resolve {base}, so the version-bump check could not run. "
            f"Fetch it first — `git fetch --depth=1 origin main:refs/remotes/"
            f"origin/main` is enough, since this diffs two trees and needs no "
            f"common ancestor — or pass --no-version deliberately."
        )
        return

    changed, paths = shipped_content_changed(base)
    if changed is None:
        fail(f"git diff against {base} failed, so the version check could not run")
        return
    if not changed:
        ok(f"no shipped plugin changes against {base} — version bump not required")
        return

    here, there = manifest_version(), manifest_version(base)
    if here is None:
        fail(f"could not read a version from {MANIFEST.relative_to(ROOT)}")
        return
    if there is None:
        ok(f"no manifest on {base} — treating as a new plugin (version {here})")
        return
    if here == there:
        files = "\n          ".join(paths)
        fail(
            f"plugin content changed but the manifest version is still {here}.\n"
            f"        The installer compares versions to decide whether to refresh "
            f"its cache, so an unchanged version means these edits reach the repo "
            f"and stop there — the marketplace clone updates, the cache that "
            f"actually executes does not (issue #2209).\n"
            f"        Bump `version` in {MANIFEST.relative_to(ROOT)}.\n"
            f"        Changed:\n          {files}"
        )
    else:
        ok(f"plugin content changed and version moved {there} -> {here}")


def main() -> int:
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--no-version", action="store_true",
                        help="skip the manifest version-bump check")
    parser.add_argument("--base", default="origin/main",
                        help="branch the version bump is measured against")
    args = parser.parse_args()

    if not HOOKS_DIR.is_dir():
        print(f"FAIL  no hooks directory at {HOOKS_DIR}")
        return 1

    check_syntax()
    check_patterns()
    check_shellcheck()
    check_fail_open()
    check_local_prior_art_needs_no_instance()
    check_session_context_reports_its_version()
    if not args.no_version:
        check_version_bump(args.base)

    print()
    if failures:
        print(f"{len(failures)} problem(s) found.")
        return 1
    print("All plugin checks passed.")
    return 0


if __name__ == "__main__":
    sys.exit(main())