@caddis/cli 0.4.51 → 0.4.53

package/bundles/antigravity-plugin/scripts/caddis_prod_check.py

@@ -0,0 +1,380 @@
+"""Check a merge on prod, item by item, from the repo's profile `.caddis/prod-check.toml`.
+
+Items (design section 10): (1) a push run exists for the merge SHA and is not cancelled; (2) the required jobs
+succeeded (with --docs-only: the deploy job was skipped); (3) the version endpoint SHA equals the merge SHA or contains
+it; (4) the health endpoint JSON has every value `true`, not only HTTP 200; (5) optional: named background jobs were
+attempted after the deploy with zero failures. A repo with no profile gets items 1 and 2 only.
+
+Prints one PASS or FAIL line per item and a RESULT line. Exit 0 only when every item passes; 1 when one fails; 3 when
+the profile cannot be read; 4 when the SHA is ambiguous or not found. Read-only: it never writes to GitHub or prod.
+
+Usage: caddis_prod_check.py [--repo PATH] [--gh-repo OWNER/NAME] [--docs-only] [--dry-run] <sha>
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import subprocess
+import sys
+import time
+import tomllib
+from dataclasses import dataclass, field
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Callable, NamedTuple
+
+try:
+    import caddis_exit
+except ModuleNotFoundError as exc:
+    if exc.name != "caddis_exit":
+        raise
+    sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "claude-harness" / "scripts"))
+    import caddis_exit
+
+FULL_SHA = re.compile(r"[0-9a-f]{40}")
+SHORT_SHA = re.compile(r"[0-9a-f]{4,39}")
+POLL_SECONDS = 30
+
+
+class Proc(NamedTuple):
+    rc: int
+    out: str
+    err: str
+
+
+Runner = Callable[[list[str]], Proc]
+
+
+class Clock:
+    def now(self) -> float:
+        return time.monotonic()
+
+    def sleep(self, seconds: float) -> None:
+        time.sleep(seconds)
+
+
+def real_runner(cwd: Path) -> Runner:
+    def run(cmd: list[str]) -> Proc:
+        try:
+            done = subprocess.run(cmd, cwd=cwd, capture_output=True, text=True, encoding="utf-8", errors="replace")
+        except FileNotFoundError as exc:
+            return Proc(127, "", str(exc))
+        return Proc(done.returncode, done.stdout, done.stderr)
+    return run
+
+
+@dataclass
+class Profile:
+    found: bool = False
+    branch: str = "main"
+    required_jobs: list[str] = field(default_factory=list)
+    deploy_job: str = "deploy_prod"
+    push_run_wait_minutes: float = 10
+    run_wait_minutes: float = 60
+    version_url: str | None = None
+    version_field: str = "sha"
+    health_url: str | None = None
+    auth: str = "none"
+    timeout_seconds: int = 60
+    background: dict | None = None
+
+
+_KEY_TYPES: dict[str, type | tuple[type, ...]] = {
+    "branch": str, "required_jobs": list, "deploy_job": str, "push_run_wait_minutes": (int, float),
+    "run_wait_minutes": (int, float), "version_url": str, "version_field": str, "health_url": str, "auth": str,
+    "timeout_seconds": int,
+}
+
+
+def load_profile(repo: Path, path: Path | None = None) -> Profile:
+    """Read `.caddis/prod-check.toml`. No file is a valid, empty profile. A bad file raises ValueError."""
+    path = path or Path(repo) / ".caddis" / "prod-check.toml"
+    if not path.is_file():
+        return Profile()
+    try:
+        data = tomllib.loads(path.read_text(encoding="utf-8"))
+    except (OSError, tomllib.TOMLDecodeError) as exc:
+        raise ValueError(f"{path}: {exc}") from exc
+    ci, prod = data.get("ci", {}), data.get("prod", {})
+    background = data.get("background")
+    if not isinstance(ci, dict) or not isinstance(prod, dict) or not isinstance(background, (dict, type(None))):
+        raise ValueError(f"{path}: [ci], [prod] and [background] must be tables")
+    profile = Profile(found=True)
+    for key, value in {**ci, **prod}.items():
+        expected = _KEY_TYPES.get(key)
+        if expected is None:
+            raise ValueError(f"{path}: unknown key {key!r}")
+        if isinstance(value, bool) or not isinstance(value, expected):
+            raise ValueError(f"{path}: {key} has the wrong type")
+        if key == "required_jobs" and not all(isinstance(job, str) for job in value):
+            raise ValueError(f"{path}: required_jobs must be a list of job names")
+        setattr(profile, key, value)
+    if profile.auth not in ("none", "ntlm"):
+        raise ValueError(f"{path}: auth must be 'none' or 'ntlm'")
+    if background is not None:
+        need = ("url", "list_key", "name_key", "attempt_key", "failure_key")
+        missing = [key for key in need if not isinstance(background.get(key), str)]
+        if missing:
+            raise ValueError(f"{path}: [background] needs {', '.join(missing)}")
+        profile.background = background
+    return profile
+
+
+@dataclass
+class Context:
+    run: Runner
+    clock: Clock
+    profile: Profile
+    repo: Path
+    gh_repo: str | None = None
+    dry_run: bool = False
+    out: Callable[[str], None] = print
+
+    def gh(self, *args: str) -> Proc:
+        cmd = ["gh", *args]
+        if self.gh_repo and args and args[0] in ("pr", "run"):
+            cmd += ["-R", self.gh_repo]
+        return self.run(cmd)
+
+    def gh_json(self, *args: str) -> object:
+        proc = self.gh(*args)
+        if proc.rc != 0:
+            raise RuntimeError(f"gh {' '.join(args[:3])} failed: {(proc.err or proc.out).strip()[:300]}")
+        return json.loads(proc.out)
+
+
+def resolve_sha(ctx: Context, sha: str) -> tuple[int, str]:
+    """Return (exit code, full SHA). A full SHA is used as given; a short one is resolved with git."""
+    sha = sha.strip().lower()
+    if FULL_SHA.fullmatch(sha):
+        return caddis_exit.CLEAN, sha
+    if not SHORT_SHA.fullmatch(sha):
+        ctx.out(f"FAIL sha {sha!r} is not a commit SHA")
+        return caddis_exit.ERROR, ""
+    for attempt in range(2):
+        proc = ctx.run(["git", "rev-parse", "--verify", "--quiet", f"{sha}^{{commit}}"])
+        if proc.rc == 0 and FULL_SHA.fullmatch(proc.out.strip()):
+            return caddis_exit.CLEAN, proc.out.strip()
+        if "ambiguous" in proc.err:
+            ctx.out(f"FAIL short sha {sha} is ambiguous; give more characters")
+            return caddis_exit.ERROR, ""
+        if attempt == 0:
+            ctx.run(["git", "fetch", "-q", "origin"])
+    ctx.out(f"FAIL sha {sha} not found, also after git fetch")
+    return caddis_exit.ERROR, ""
+
+
+def push_runs(ctx: Context, limit: int = 20) -> list[dict]:
+    return ctx.gh_json("run", "list", "--branch", ctx.profile.branch, "--event", "push", "--limit", str(limit),
+                       "--json", "databaseId,headSha,status,conclusion")  # type: ignore[return-value]
+
+
+def find_push_run(ctx: Context, sha: str) -> dict | None:
+    """Wait up to the profile's minutes for a push run of `sha` to appear."""
+    deadline = ctx.clock.now() + ctx.profile.push_run_wait_minutes * 60
+    while True:
+        run = next((r for r in push_runs(ctx) if r.get("headSha") == sha), None)
+        if run or ctx.dry_run or ctx.clock.now() >= deadline:
+            return run
+        ctx.clock.sleep(15)
+
+
+def wait_run(ctx: Context, run_id: int) -> dict | None:
+    """Return the completed run view (status, conclusion, jobs), or None when it does not complete in time."""
+    deadline = ctx.clock.now() + ctx.profile.run_wait_minutes * 60
+    while True:
+        view = ctx.gh_json("run", "view", str(run_id), "--json", "status,conclusion,jobs,headSha")
+        if view.get("status") == "completed" or ctx.dry_run:  # type: ignore[union-attr]
+            return view  # type: ignore[return-value]
+        if ctx.clock.now() >= deadline:
+            return None
+        ctx.clock.sleep(POLL_SECONDS)
+
+
+def _jobs(view: dict) -> dict[str, str]:
+    return {j.get("name", ""): (j.get("conclusion") or j.get("status") or "") for j in view.get("jobs", [])}
+
+
+def _deploy_done(view: dict, deploy_job: str) -> datetime | None:
+    for job in view.get("jobs", []):
+        if job.get("name") == deploy_job and job.get("completedAt"):
+            return _utc(job["completedAt"])
+    return None
+
+
+def _utc(text: str) -> datetime:
+    when = datetime.fromisoformat(text.replace("Z", "+00:00"))
+    return when if when.tzinfo else when.replace(tzinfo=timezone.utc)
+
+
+def check_ci(ctx: Context, sha: str, docs_only: bool) -> tuple[list[str], dict | None]:
+    """Items 1 and 2. Return (failed item labels, completed run view)."""
+    run = find_push_run(ctx, sha)
+    if not run:
+        ctx.out(f"FAIL 1 no push run for {sha} on {ctx.profile.branch} after "
+                f"{ctx.profile.push_run_wait_minutes:g} min")
+        return ["1 push run"], None
+    if ctx.dry_run and run.get("status") != "completed":
+        ctx.out(f"INFO 1 push run {run['databaseId']} is {run.get('status')}; would wait for it")
+        return [], None
+    view = wait_run(ctx, run["databaseId"])
+    if view is None:
+        ctx.out(f"FAIL 1 push run {run['databaseId']} did not complete in {ctx.profile.run_wait_minutes:g} min")
+        return ["1 push run"], None
+    conclusion = view.get("conclusion") or ""
+    if conclusion in ("cancelled", ""):
+        ctx.out(f"FAIL 1 push run {run['databaseId']} conclusion {conclusion or 'none'}")
+        return ["1 push run"], view
+    ctx.out(f"PASS 1 push run {run['databaseId']} completed ({conclusion})")
+    jobs = _jobs(view)
+    ctx.out("jobs: " + ", ".join(f"{name}={state}" for name, state in jobs.items()))
+    deploy = ctx.profile.deploy_job
+    if docs_only:
+        if jobs.get(deploy) == "skipped" and conclusion == "success":
+            ctx.out(f"PASS 2 docs-only: {deploy} skipped, run conclusion success")
+            return [], view
+        if conclusion != "success":
+            ctx.out(f"FAIL 2 docs-only merge, but the run concluded {conclusion}")
+            return ["2 jobs"], view
+        ctx.out(f"FAIL 2 docs-only merge, but {deploy} = {jobs.get(deploy, 'absent')}; run the full prod check")
+        return ["2 jobs"], view
+    if not ctx.profile.found:
+        if conclusion == "success":
+            ctx.out("PASS 2 run conclusion success (no profile: required jobs unknown)")
+            return [], view
+        ctx.out(f"FAIL 2 run conclusion {conclusion}")
+        return ["2 jobs"], view
+    bad = {name: jobs.get(name, "absent") for name in ctx.profile.required_jobs if jobs.get(name) != "success"}
+    if deploy and jobs.get(deploy) != "success":
+        # A code merge that did not deploy fails, also when the job is not listed as required.
+        # A repo with no deploy job sets `deploy_job = ""` in its profile.
+        bad[deploy] = jobs.get(deploy, "absent")
+    if bad:
+        skipped = " (skipped without --docs-only)" if bad.get(deploy) == "skipped" else ""
+        ctx.out("FAIL 2 " + ", ".join(f"{n}={c}" for n, c in bad.items()) + skipped)
+        return ["2 jobs"], view
+    ctx.out(f"PASS 2 {', '.join(ctx.profile.required_jobs) or 'no required jobs'} success")
+    return [], view
+
+
+def curl(ctx: Context, url: str) -> tuple[int, object]:
+    """GET `url`. Return (HTTP code, parsed JSON or None)."""
+    auth = ["--ntlm", "-u", ":"] if ctx.profile.auth == "ntlm" else []
+    proc = ctx.run(["curl", "-s", "-m", str(ctx.profile.timeout_seconds), *auth, "-w", "\n%{http_code}", url])
+    body, _, code = proc.out.rpartition("\n")
+    try:
+        status = int(code.strip())
+    except ValueError:
+        status = 0
+    try:
+        return status, json.loads(body)
+    except ValueError:
+        return status, None
+
+
+def _false_leaves(value: object, path: str = "") -> list[str]:
+    """Paths whose leaf value is not exactly True. An empty object counts as not healthy."""
+    if isinstance(value, dict):
+        if not value:
+            return [path or "(empty)"]
+        return [p for key, sub in value.items() for p in _false_leaves(sub, f"{path}.{key}" if path else str(key))]
+    return [] if value is True else [f"{path}={value!r}"]
+
+
+def check_version(ctx: Context, sha: str) -> list[str]:
+    status, body = curl(ctx, ctx.profile.version_url or "")
+    found = body.get(ctx.profile.version_field) if isinstance(body, dict) else None
+    if status == 200 and isinstance(found, str) and found:
+        if found == sha:
+            ctx.out(f"PASS 3 version sha {found}")
+            return []
+        if ctx.run(["git", "merge-base", "--is-ancestor", sha, found]).rc == 0:
+            ctx.out(f"PASS 3 version sha {found} contains {sha}")
+            return []
+    ctx.out(f"FAIL 3 version sha {found!r} (HTTP {status}) is not {sha} and does not contain it")
+    return ["3 version"]
+
+
+def check_health(ctx: Context) -> list[str]:
+    status, body = curl(ctx, ctx.profile.health_url or "")
+    bad = _false_leaves(body) if isinstance(body, dict) else ["body is not a JSON object"]
+    if status == 200 and not bad:
+        ctx.out(f"PASS 4 health {json.dumps(body)}")
+        return []
+    ctx.out(f"FAIL 4 health HTTP {status}; not true: {', '.join(bad) or '-'}")
+    return ["4 health"]
+
+
+def check_background(ctx: Context, since: datetime | None) -> list[str]:
+    spec = ctx.profile.background or {}
+    since = since or datetime.now(timezone.utc)
+    deadline = ctx.clock.now() + float(spec.get("wait_minutes", 15)) * 60
+    while True:
+        status, body = curl(ctx, spec["url"])
+        jobs = body.get(spec["list_key"]) if isinstance(body, dict) and status == 200 else None
+        if not isinstance(jobs, list) or not jobs:
+            ctx.out(f"FAIL 5 background list {spec['list_key']!r} missing or empty (HTTP {status})")
+            return ["5 background"]
+        failed = [(j.get(spec["name_key"]), j.get(spec["failure_key"])) for j in jobs if j.get(spec["failure_key"])]
+        late = [j.get(spec["name_key"]) for j in jobs
+                if not j.get(spec["attempt_key"]) or _utc(str(j[spec["attempt_key"]])) < since]
+        if failed:
+            ctx.out(f"FAIL 5 failed background jobs: {failed}")
+            return ["5 background"]
+        if not late:
+            ctx.out(f"PASS 5 {len(jobs)} background jobs attempted after {since:%H:%M:%S} UTC, no failures")
+            return []
+        if ctx.clock.now() >= deadline:
+            ctx.out(f"FAIL 5 not attempted since the deploy: {late}")
+            return ["5 background"]
+        ctx.clock.sleep(60)
+
+
+def check(ctx: Context, sha: str, docs_only: bool = False) -> int:
+    code, full = resolve_sha(ctx, sha)
+    if code != caddis_exit.CLEAN:
+        return code
+    if not ctx.profile.found:
+        ctx.out("no profile .caddis/prod-check.toml: checking items 1 and 2 only")
+    try:
+        fails, view = check_ci(ctx, full, docs_only)
+        if not fails and not docs_only and ctx.profile.found:
+            if ctx.dry_run:
+                ctx.out("would check items 3 to 5 against prod (version, health, background jobs)")
+            else:
+                if ctx.profile.version_url:
+                    fails += check_version(ctx, full)
+                if ctx.profile.health_url:
+                    fails += check_health(ctx)
+                if ctx.profile.background:
+                    fails += check_background(ctx, _deploy_done(view or {}, ctx.profile.deploy_job))
+    except (RuntimeError, ValueError, KeyError) as exc:
+        ctx.out(f"FAIL could not read CI state: {exc}")
+        fails = ["CI state"]
+    ctx.out("RESULT PASS" if not fails else "RESULT FAIL " + ", ".join(fails))
+    return caddis_exit.CLEAN if not fails else caddis_exit.BLOCKED
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__.splitlines()[0])
+    parser.add_argument("sha")
+    parser.add_argument("--repo", type=Path, default=Path("."))
+    parser.add_argument("--gh-repo", help="OWNER/NAME when the repo's remote is not the one to read")
+    parser.add_argument("--profile", type=Path, help="default: <repo>/.caddis/prod-check.toml")
+    parser.add_argument("--docs-only", action="store_true", help="the merge changed docs only: expect no deploy")
+    parser.add_argument("--dry-run", action="store_true", help="read CI state only; never call prod")
+    args = parser.parse_args(argv)
+    try:
+        profile = load_profile(args.repo, args.profile)
+    except ValueError as exc:
+        print(f"bad profile: {exc}", file=sys.stderr)
+        return caddis_exit.NOT_RUN
+    ctx = Context(run=real_runner(args.repo), clock=Clock(), profile=profile, repo=args.repo,
+                  gh_repo=args.gh_repo, dry_run=args.dry_run)
+    return check(ctx, args.sha, docs_only=args.docs_only)
+
+
+if __name__ == "__main__":
+    sys.exit(main())

package/bundles/antigravity-plugin/scripts/caddis_redact.py

@@ -0,0 +1,99 @@
+#!/usr/bin/env python3
+"""caddis_redact - keep provider keys out of caddis logs and captured output.
+
+Why: on 10 Oct 2026 a review launcher session printed the GLM API key into its local session log while
+debugging (caddis-next proposal, issue 4); the owner had to rotate the key. caddis captures model and
+launcher output into review files, lane packets and printed summaries. Every such text goes through
+`redact()` (or `redact_obj()` for a packet) first.
+
+Secrets are the VALUES (never the names) of:
+- every `NAME=value` line in the keys file whose NAME looks secret (contains KEY, TOKEN, SECRET, PASS,
+  AUTH or CREDENTIAL): `$CADDIS_KEYS_FILE` if set (with `~` expanded, as oss_model and oss_review do),
+  else `~/.caddis/keys.env`; optional `export `, quotes stripped the way the key consumers strip them;
+- the same kind of variables in the environment (`REVIEW_API_KEY`, `OSS_API_KEY`, `<P>_API_KEY`, which
+  the consumers prefer over the file).
+Values shorter than 8 characters are ignored. Each value is also masked in its JSON-escaped form, so a key
+with `\\`, `"` or `<>&` (Go-style `\\u003c`) does not survive inside raw JSON output.
+
+Limit: only whole, exact copies are masked. A key split across lines, cut short (`sk-abc...xyz`) or
+re-encoded (base64) passes.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import re
+from pathlib import Path
+
+MIN_SECRET_LEN = 8
+MASK = "***"
+_SENSITIVE_NAME = re.compile(r"KEY|TOKEN|SECRET|PASS|AUTH|CREDENTIAL", re.I)
+
+
+def keys_file(home: Path) -> Path:
+    override = os.environ.get("CADDIS_KEYS_FILE")
+    return Path(override).expanduser() if override else home / ".caddis" / "keys.env"
+
+
+def _forms(value: str) -> set[str]:
+    """The value and the forms it takes inside raw JSON output (standard and Go-style escapes)."""
+    forms = {value}
+    escaped = json.dumps(value)[1:-1]
+    forms.add(escaped)
+    forms.add(escaped.replace("<", "\\u003c").replace(">", "\\u003e").replace("&", "\\u0026"))
+    return forms
+
+
+def load_secrets(home: Path) -> list[str]:
+    """The secret values in every form to mask, longest first (so a key that contains another is masked whole)."""
+    values: set[str] = set()
+    try:
+        text = keys_file(home).read_text(encoding="utf-8-sig", errors="replace")
+    except OSError:
+        text = ""
+    for raw in text.splitlines():
+        line = raw.strip()
+        if not line or line.startswith("#") or "=" not in line:
+            continue
+        if line.startswith("export "):
+            line = line[len("export "):].lstrip()
+        name, value = line.split("=", 1)
+        if not _SENSITIVE_NAME.search(name):
+            continue
+        value = value.strip()
+        # The consumers (oss_model, oss_review) use .strip('"').strip("'"); mask that value AND the
+        # balanced-quote value, so an unbalanced quote cannot leave the key that is really used unmasked.
+        for candidate in {value.strip('"').strip("'"),
+                          value[1:-1] if len(value) >= 2 and value[0] == value[-1] and value[0] in "'\"" else value}:
+            if len(candidate) >= MIN_SECRET_LEN:
+                values.add(candidate)
+    for name, value in os.environ.items():
+        if (_SENSITIVE_NAME.search(name) and len(value) >= MIN_SECRET_LEN and name not in ("CADDIS_KEYS_FILE",)
+                and not name.startswith("GIT_CONFIG_")):  # GIT_CONFIG_KEY_<n> holds a setting NAME, not a secret
+            values.add(value)
+    forms: set[str] = set()
+    for value in values:
+        forms |= _forms(value)
+    return sorted(forms, key=len, reverse=True)
+
+
+def redact(text: str, secrets: list[str]) -> str:
+    """`text` with every secret form replaced by ***."""
+    if not text:
+        return text
+    for secret in secrets:
+        if secret in text:
+            text = text.replace(secret, MASK)
+    return text
+
+
+def redact_obj(obj, secrets: list[str]):
+    """A copy of a JSON-able packet (dict/list/str) with every string redacted."""
+    if isinstance(obj, str):
+        return redact(obj, secrets)
+    if isinstance(obj, dict):
+        return {k: redact_obj(v, secrets) for k, v in obj.items()}
+    if isinstance(obj, list):
+        return [redact_obj(v, secrets) for v in obj]
+    return obj

package/bundles/antigravity-plugin/scripts/caddis_relay_git.py

@@ -124,8 +124,8 @@ def fetch_relay(repo: Path, workstream: str, *, replace: bool = False) -> str:
 
 
 def _headless() -> bool:
-    """No human is watching: the caddis launchers and the docket runner set these."""
-    return any(os.environ.get(name) for name in ("CADDIS_HEADLESS", "DOCKET_PLAN", "DOCKET_BRANCH"))
+    """No human is watching: the caddis launchers set CADDIS_HEADLESS."""
+    return bool(os.environ.get("CADDIS_HEADLESS"))
 
 
 def main(argv: list[str] | None = None) -> int: