atlas-iac/scripts/ops/hermes_handoff_arming.py

155 lines
5.6 KiB
Python
Raw Normal View History

fix(hermes): close the zero-evidence fail-open in absence checks evaluate_names_absent returned PASS when its step exited 0 with no output, so five mandatory checks - the ones asserting that provider API keys, forge credentials, a cluster-admin binding, and shared coordinator state are absent - could report a pass on no evidence and turn a NO_GO into a GO. Both name rules now resolve their step through one guard in _line_step, so zero observations are NOT_RUN. Regressions pin all five real catalog specs plus both reachable silence paths: a POSIX pipeline whose status comes from its last stage, and a drifted kubectl -o jsonpath. The pool claim projection emits one <volume>=<claim> line per template volume so a volume without a PVC still counts as an observation rather than reading as drift. Also closes the review's reachable hardening and evidence defects: - pin Gitea paths to atlas/titan-iac on an exact segment boundary and reject relative segments, including percent-encoded ones - forbid impersonation structurally in every mode and vantage; the inner command of kubectl exec is re-checked rather than exempted, and validate_catalog no longer guards only the operator vantage - drop flux and helm from the binary allowlist; they had no pinned release digest, so no allowlisted binary can now be admitted that the executor would refuse to attest - remove the inert --concurrency and --expect-telegram-sessions flags and the dead concurrency bound; Telegram continuity stays mandatory - read the ephemeral pull index page by page, treat the create response as an authoritative source for the pull number, close every number either source names, and surface residue_ref plus exact manual_cleanup commands when creation is uncertain - keep executable_path and executable_sha256 on unrecorded bulk-evidence steps so withholding bytes never withholds binary attestation - revert the repo-wide hygiene legacy-exception mechanism; the contract change here is purely additive and the four pre-existing over-cap files are left to the canonical contract change in PR #14/#15 - correct the runbook ruff format scope so the documented command passes Split hermes_handoff_arming.py out of hermes_handoff_ephemeral.py to keep both modules under the 500-line cap. All 16 handoff modules hold at least 95% line and branch coverage; the mutation gate is 13/13. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-17 16:24:37 +00:00
#!/usr/bin/env python3
"""Arming, provenance, and push-target rules for the armed ephemeral mode.
Nothing here touches the network or spawns a process. It is the half of armed
mode that must be true before a single command exists: the confirmation phrase,
the fixed repository and base, the one ephemeral ref grammar, and a local
worktree whose HEAD and ``origin/main`` are byte-exact with the reviewed release.
"""
from __future__ import annotations
import configparser
import re
from dataclasses import dataclass
from pathlib import Path
from hermes_handoff_policy import (
EXPECTED_BASE,
EXPECTED_REMOTE,
EXPECTED_REPO,
arm_ephemeral_policy,
)
CONFIRMATION = "ARM EPHEMERAL HERMES HANDOFF PUSH"
REF_PREFIX = "ephemeral/hermes-handoff-acceptance"
REF_RE = re.compile(rf"{REF_PREFIX}/[a-z0-9][a-z0-9-]{{7,63}}\Z")
TOKEN_RE = re.compile(r"[a-z0-9][a-z0-9-]{7,63}\Z")
SHA_RE = re.compile(r"[0-9a-f]{40}\Z")
EXPECTED_HEAD_REF = "feature/hermes-full-handoff-acceptance"
EXPECTED_REMOTE_URL = "https://scm.bstein.dev/titan/atlas-iac.git"
fix(hermes): close the zero-evidence fail-open in absence checks evaluate_names_absent returned PASS when its step exited 0 with no output, so five mandatory checks - the ones asserting that provider API keys, forge credentials, a cluster-admin binding, and shared coordinator state are absent - could report a pass on no evidence and turn a NO_GO into a GO. Both name rules now resolve their step through one guard in _line_step, so zero observations are NOT_RUN. Regressions pin all five real catalog specs plus both reachable silence paths: a POSIX pipeline whose status comes from its last stage, and a drifted kubectl -o jsonpath. The pool claim projection emits one <volume>=<claim> line per template volume so a volume without a PVC still counts as an observation rather than reading as drift. Also closes the review's reachable hardening and evidence defects: - pin Gitea paths to atlas/titan-iac on an exact segment boundary and reject relative segments, including percent-encoded ones - forbid impersonation structurally in every mode and vantage; the inner command of kubectl exec is re-checked rather than exempted, and validate_catalog no longer guards only the operator vantage - drop flux and helm from the binary allowlist; they had no pinned release digest, so no allowlisted binary can now be admitted that the executor would refuse to attest - remove the inert --concurrency and --expect-telegram-sessions flags and the dead concurrency bound; Telegram continuity stays mandatory - read the ephemeral pull index page by page, treat the create response as an authoritative source for the pull number, close every number either source names, and surface residue_ref plus exact manual_cleanup commands when creation is uncertain - keep executable_path and executable_sha256 on unrecorded bulk-evidence steps so withholding bytes never withholds binary attestation - revert the repo-wide hygiene legacy-exception mechanism; the contract change here is purely additive and the four pre-existing over-cap files are left to the canonical contract change in PR #14/#15 - correct the runbook ruff format scope so the documented command passes Split hermes_handoff_arming.py out of hermes_handoff_ephemeral.py to keep both modules under the 500-line cap. All 16 handoff modules hold at least 95% line and branch coverage; the mutation gate is 13/13. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-17 16:24:37 +00:00
PROTECTED_NAMES = frozenset(
{
"default",
"develop",
"head",
"main",
"master",
"prod",
"production",
"release",
"stable",
"trunk",
}
)
class ArmingError(ValueError):
"""Raised when an ephemeral run is not safe to start."""
@dataclass(frozen=True)
class ArmRequest:
"""Every fixed and caller-confirmed input to one armed run."""
repo: str
remote: str
token: str
confirmation: str
base: str = EXPECTED_BASE
expected_head: str = ""
expected_base_sha: str = ""
@property
def ref(self) -> str:
return f"{REF_PREFIX}/{self.token}"
def normalise_branch(name: str) -> str:
"""Return a branch name stripped of ref decoration, for comparison."""
return name.strip().removeprefix("refs/heads/").strip("/").lower()
def assert_push_target_allowed(ref: str) -> None:
"""Raise unless ``ref`` is the one ephemeral branch grammar."""
if normalise_branch(ref) in PROTECTED_NAMES:
raise ArmingError(f"refusing to push to protected branch {ref!r}")
if not REF_RE.fullmatch(ref):
raise ArmingError(f"push target {ref!r} is not an ephemeral acceptance ref")
def bounded_read(path: Path, limit: int = 1024 * 1024) -> str:
data = path.read_bytes()
if not data or len(data) > limit:
raise ArmingError(f"local Git metadata is empty or exceeds {limit} bytes")
return data.decode("utf-8", "strict")
def _attest_worktree(root: Path, expected_head: str, expected_base_sha: str) -> None:
"""Bind cwd, common Git metadata, branch, remote, and exact local HEAD."""
resolved = root.resolve(strict=True)
marker = resolved / ".git"
if not marker.is_file():
raise ArmingError(
"armed mode must run at the root of the existing linked worktree"
)
line = bounded_read(marker, 4096).strip()
if not line.startswith("gitdir: "):
raise ArmingError("linked-worktree Git metadata is malformed")
gitdir = Path(line.removeprefix("gitdir: ")).resolve(strict=True)
common = (gitdir / bounded_read(gitdir / "commondir", 4096).strip()).resolve(
strict=True
)
head = bounded_read(gitdir / "HEAD", 4096).strip()
expected_ref = f"refs/heads/{EXPECTED_HEAD_REF}"
if head != f"ref: {expected_ref}":
raise ArmingError("armed mode must run from the existing PR #19 feature branch")
ref_path = common / expected_ref
actual_head = bounded_read(ref_path, 4096).strip() if ref_path.is_file() else ""
if actual_head != expected_head:
raise ArmingError(
"local worktree HEAD does not match the exact reviewed PR head"
)
remote_ref = common / "refs" / "remotes" / EXPECTED_REMOTE / EXPECTED_BASE
remote_main = bounded_read(remote_ref, 4096).strip() if remote_ref.is_file() else ""
if remote_main != expected_base_sha:
raise ArmingError("local origin/main does not match the exact release base SHA")
parser = configparser.RawConfigParser(interpolation=None)
try:
parser.read_string(bounded_read(common / "config"))
remote_url = parser.get(f'remote "{EXPECTED_REMOTE}"', "url")
except (configparser.Error, KeyError) as exc:
raise ArmingError("local Git remote metadata is malformed") from exc
if remote_url != EXPECTED_REMOTE_URL:
raise ArmingError("origin is not the fixed Titan atlas-iac HTTPS repository")
fix(hermes): close the zero-evidence fail-open in absence checks evaluate_names_absent returned PASS when its step exited 0 with no output, so five mandatory checks - the ones asserting that provider API keys, forge credentials, a cluster-admin binding, and shared coordinator state are absent - could report a pass on no evidence and turn a NO_GO into a GO. Both name rules now resolve their step through one guard in _line_step, so zero observations are NOT_RUN. Regressions pin all five real catalog specs plus both reachable silence paths: a POSIX pipeline whose status comes from its last stage, and a drifted kubectl -o jsonpath. The pool claim projection emits one <volume>=<claim> line per template volume so a volume without a PVC still counts as an observation rather than reading as drift. Also closes the review's reachable hardening and evidence defects: - pin Gitea paths to atlas/titan-iac on an exact segment boundary and reject relative segments, including percent-encoded ones - forbid impersonation structurally in every mode and vantage; the inner command of kubectl exec is re-checked rather than exempted, and validate_catalog no longer guards only the operator vantage - drop flux and helm from the binary allowlist; they had no pinned release digest, so no allowlisted binary can now be admitted that the executor would refuse to attest - remove the inert --concurrency and --expect-telegram-sessions flags and the dead concurrency bound; Telegram continuity stays mandatory - read the ephemeral pull index page by page, treat the create response as an authoritative source for the pull number, close every number either source names, and surface residue_ref plus exact manual_cleanup commands when creation is uncertain - keep executable_path and executable_sha256 on unrecorded bulk-evidence steps so withholding bytes never withholds binary attestation - revert the repo-wide hygiene legacy-exception mechanism; the contract change here is purely additive and the four pre-existing over-cap files are left to the canonical contract change in PR #14/#15 - correct the runbook ruff format scope so the documented command passes Split hermes_handoff_arming.py out of hermes_handoff_ephemeral.py to keep both modules under the 500-line cap. All 16 handoff modules hold at least 95% line and branch coverage; the mutation gate is 13/13. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-17 16:24:37 +00:00
def preflight(
request: ArmRequest, expected_repo: str, worktree: Path | None = None
) -> None:
"""Validate arming and local provenance before any subprocess/network call."""
if request.confirmation != CONFIRMATION:
raise ArmingError("confirmation phrase does not match")
if expected_repo != EXPECTED_REPO or request.repo != EXPECTED_REPO:
raise ArmingError(
"armed repository is not the fixed Titan atlas-iac repository"
fix(hermes): close the zero-evidence fail-open in absence checks evaluate_names_absent returned PASS when its step exited 0 with no output, so five mandatory checks - the ones asserting that provider API keys, forge credentials, a cluster-admin binding, and shared coordinator state are absent - could report a pass on no evidence and turn a NO_GO into a GO. Both name rules now resolve their step through one guard in _line_step, so zero observations are NOT_RUN. Regressions pin all five real catalog specs plus both reachable silence paths: a POSIX pipeline whose status comes from its last stage, and a drifted kubectl -o jsonpath. The pool claim projection emits one <volume>=<claim> line per template volume so a volume without a PVC still counts as an observation rather than reading as drift. Also closes the review's reachable hardening and evidence defects: - pin Gitea paths to atlas/titan-iac on an exact segment boundary and reject relative segments, including percent-encoded ones - forbid impersonation structurally in every mode and vantage; the inner command of kubectl exec is re-checked rather than exempted, and validate_catalog no longer guards only the operator vantage - drop flux and helm from the binary allowlist; they had no pinned release digest, so no allowlisted binary can now be admitted that the executor would refuse to attest - remove the inert --concurrency and --expect-telegram-sessions flags and the dead concurrency bound; Telegram continuity stays mandatory - read the ephemeral pull index page by page, treat the create response as an authoritative source for the pull number, close every number either source names, and surface residue_ref plus exact manual_cleanup commands when creation is uncertain - keep executable_path and executable_sha256 on unrecorded bulk-evidence steps so withholding bytes never withholds binary attestation - revert the repo-wide hygiene legacy-exception mechanism; the contract change here is purely additive and the four pre-existing over-cap files are left to the canonical contract change in PR #14/#15 - correct the runbook ruff format scope so the documented command passes Split hermes_handoff_arming.py out of hermes_handoff_ephemeral.py to keep both modules under the 500-line cap. All 16 handoff modules hold at least 95% line and branch coverage; the mutation gate is 13/13. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-17 16:24:37 +00:00
)
if (
request.remote != EXPECTED_REMOTE
or normalise_branch(request.base) != EXPECTED_BASE
):
raise ArmingError("armed remote and pull-request base are fixed to origin/main")
if not TOKEN_RE.fullmatch(request.token):
raise ArmingError(
"ephemeral token must be 8-64 lowercase alphanumeric or dash characters"
)
if not SHA_RE.fullmatch(request.expected_head) or not SHA_RE.fullmatch(
request.expected_base_sha
):
raise ArmingError("exact reviewed head and release base SHAs are required")
assert_push_target_allowed(request.ref)
_attest_worktree(
worktree or Path.cwd(), request.expected_head, request.expected_base_sha
)
arm_ephemeral_policy(request.ref)