atlas-iac/scripts/ops/hermes_handoff_checks_workers.py
Hermes Agent d208a89d9d 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

377 lines
13 KiB
Python

#!/usr/bin/env python3
"""Acceptance checks for the distributed worker pool and the visible surfaces.
A pool is only distributed if its workers occupy distinct nodes and distinct
volumes and leave state ownership with the coordinator; a replica count alone
satisfies none of that. A surface is only working if the work it did is visible
in it, which is why the session and Kanban checks look at the same data the UI
renders rather than at whether a pod is Running.
The ephemeral entries are placeholders in a default run. They report NOT_RUN
until the mutation mode is separately armed, and the armed mode replaces them
with real, mandatory results.
"""
from __future__ import annotations
from hermes_handoff_catalog import (
CHAT,
INIT_NAME_PROJECTION,
OPERATOR,
POOL_PROJECTION,
SELF,
Targets,
check,
)
from hermes_handoff_catalog import step as step
from hermes_handoff_model import EPHEMERAL, CheckSpec
from hermes_handoff_policy import projection
KANBAN_STATUSES = [
"archived",
"blocked",
"done",
"ready",
"review",
"running",
"scheduled",
"todo",
"triage",
]
def _pool(targets: Targets) -> list[CheckSpec]:
selector = targets.pool_selector
return [
check(
"pool.three-workers-have-distinct-nodes-and-volumes",
"Exactly three Ready ordinal workers occupy distinct nodes and claims",
"pool",
"pool_topology",
[
step(
"statefulset",
OPERATOR,
"kubectl",
"--namespace",
targets.namespace,
"get",
f"statefulset/{targets.pool_statefulset}",
"-o",
projection(
"jsonpath={.spec.replicas}{'\\t'}{.status.readyReplicas}{'\\t'}{.status.currentReplicas}{'\\t'}{.spec.template.spec.automountServiceAccountToken}".replace(
"'", '"'
)
),
),
step(
"workers",
OPERATOR,
"kubectl",
"--namespace",
targets.namespace,
"get",
"pods",
"--selector",
selector,
"-o",
POOL_PROJECTION,
),
],
{
"state_step": "statefulset",
"worker_step": "workers",
"replicas": targets.pool_replicas,
"name": targets.pool_statefulset,
},
rationale="Replica count, Ready state, ordinal coverage, node placement, PVC isolation, and tokenlessness are one fail-closed assertion.",
),
check(
"pool.coordinator-retains-sole-state-ownership",
"No pool worker mounts the coordinator's durable state",
"pool",
"names_absent",
[
step(
"claims",
OPERATOR,
"kubectl",
"--namespace",
targets.namespace,
"get",
f"statefulset/{targets.pool_statefulset}",
"-o",
projection(
"jsonpath={range .spec.template.spec.volumes[*]}{.name}{'='}"
"{.persistentVolumeClaim.claimName}{'\\n'}{end}".replace(
"'", '"'
)
),
)
],
{"step": "claims", "contains": targets.coordinator_claims},
rationale="Two writers on one SQLite state file is the duplicate-work and lost-result failure this pool exists to avoid. One line per template volume keeps a drifted or empty projection NOT_RUN instead of a silent pass.",
),
*_pool_env(targets),
]
def _pool_env(targets: Targets) -> list[CheckSpec]:
"""Return the mandatory assignment-safety configuration check."""
return [
check(
"pool.assignment-safety-knobs-are-configured",
"Pool workers carry the lease and de-duplication configuration",
"pool",
"names_present",
[
step(
"env",
OPERATOR,
"kubectl",
"--namespace",
targets.namespace,
"get",
f"statefulset/{targets.pool_statefulset}",
"-o",
projection(
'jsonpath={range .spec.template.spec.containers[*]}{range .env[*]}{.name}{"\\n"}{end}{end}'
),
)
],
{"step": "env", "names": targets.pool_worker_env},
rationale="Deployed configuration only; the stale-claim and duplicate-dispatch behaviour itself is covered by the pool's own regression suite.",
),
check(
"pool.assignment-fieldrefs-are-bound",
"Worker node and ordinal identities come from fixed downward-API fields",
"pool",
"names_present",
[
step(
"source",
OPERATOR,
"git",
"show",
f"{targets.remote}/main:services/hermes/execution-worker-statefulset.yaml",
record=False,
)
],
{
"step": "source",
"names": (
"name: HERMES_WORKER_ORDINAL",
"fieldPath: metadata.labels['apps.kubernetes.io/pod-index']",
"name: HERMES_WORKER_NODE",
"fieldPath: spec.nodeName",
),
},
rationale="This prevents caller-authored ordinal/node values from impersonating another worker assignment.",
),
]
def _surfaces(targets: Targets) -> list[CheckSpec]:
checks = [
check(
"surfaces.kanban-activity-is-visible",
"Durable Kanban activity is visible to the agent surface",
"surfaces",
"all_of_field",
[
step(
"tasks",
SELF,
"hermes",
"kanban",
"list",
"--json",
record=False,
max_bytes=targets.listing_bytes,
)
],
{"step": "tasks", "key_field": "id", "fields": {"status": KANBAN_STATUSES}},
rationale="Non-empty and well-formed. An empty board is NOT_RUN, because it cannot distinguish a quiet week from a broken read.",
),
check(
"surfaces.durable-sessions-render-without-blank-rows",
"Durable sessions are listed with a preview, a source, and an id",
"surfaces",
"lines_match",
[
step(
"sessions",
SELF,
"hermes",
"sessions",
"list",
"--limit",
"20",
record=False,
)
],
{"step": "sessions", "pattern": r"^\S.*\S$", "skip": 2, "minimum": 1},
rationale="A row whose leading column is blank is the session-render regression seen from the data side.",
),
check(
"surfaces.agent-ui-session-patches-are-deployed",
"The agent surface runs the session-activity rendering patches",
"surfaces",
"names_present",
[
step(
"init",
OPERATOR,
"kubectl",
"--namespace",
targets.namespace,
"get",
f"deploy/{targets.agent_deployment}",
"-o",
INIT_NAME_PROJECTION,
)
],
{
"step": "init",
"names": ("patch-web-session-activity", "patch-api-server-sessions"),
},
rationale="Blank session rows were a rendering regression; this asserts the patches that fixed it are in the running pod.",
),
check(
"surfaces.chat-session-patches-are-deployed",
"The chat surface runs the API-session continuity patches",
"surfaces",
"names_present",
[
step(
"init",
OPERATOR,
"kubectl",
"--namespace",
targets.namespace,
"get",
f"statefulset/{targets.chat_statefulset}",
"-o",
INIT_NAME_PROJECTION,
)
],
{"step": "init", "names": targets.chat_init_containers},
),
check(
"surfaces.chat-telegram-topic-state-is-durable",
"Telegram topic and tenant state is held on durable storage",
"surfaces",
"names_present",
[
step(
"claims",
OPERATOR,
"kubectl",
"--namespace",
targets.namespace,
"get",
f"deploy/{targets.chat_router_deployment}",
"-o",
projection(
'jsonpath={range .spec.template.spec.volumes[*]}{.persistentVolumeClaim.claimName}{"\\n"}{end}'
),
)
],
{"step": "claims", "contains": ("state",)},
rationale="Topic selection survives a router restart only if the tenant state file is on a claim rather than in the pod.",
),
]
checks.append(
check(
"surfaces.chat-telegram-sessions-are-continuous",
"The selected chat tenant still holds durable Telegram-sourced sessions",
"surfaces",
"lines_match",
[
step(
"telegram",
CHAT,
"hermes",
"sessions",
"list",
"--source",
"telegram",
"--limit",
"20",
record=False,
)
],
{"step": "telegram", "pattern": r"^\S.*\S$", "skip": 2, "minimum": 1},
rationale="Session continuity is per tenant and depends on there having been Telegram traffic, so it is asserted only when the operator names the tenant that carries the identity.",
)
)
checks.append(
check(
"surfaces.chat-runs-the-telegram-topic-revision",
"The chat surface runs the durable Telegram topic configuration",
"surfaces",
"stdout_matches",
[
step(
"revision",
OPERATOR,
"kubectl",
"--namespace",
targets.namespace,
"get",
f"statefulset/{targets.chat_statefulset}",
"-o",
projection(
"jsonpath={.spec.template.metadata.annotations.ai\\.bstein\\.dev/config-rev}"
),
)
],
{"step": "revision", "equals": targets.chat_config_revision},
rationale="Telegram topic continuity is carried by the config revision the pod template pins.",
)
)
return checks
def _ephemeral() -> list[CheckSpec]:
return [
check(
f"ephemeral.{name}",
title,
"ephemeral",
"not_armed",
[],
{"reason": reason},
mandatory=False,
scope=EPHEMERAL,
rationale="Mutating checks are separately armed. A default run reports them NOT_RUN and never touches the forge.",
)
for name, title, reason in (
(
"feature-branch-push",
"A unique ephemeral feature branch can be pushed and removed",
"ephemeral mutation mode is not armed",
),
(
"draft-pull-request",
"A draft pull request can be opened and closed against the ephemeral branch",
"ephemeral mutation mode is not armed",
),
(
"protected-branch-refusal",
"Arming against a protected branch is refused before any network call",
"ephemeral mutation mode is not armed",
),
(
"cleanup-verified",
"Every ephemeral ref and pull request is removed and the removal is verified",
"ephemeral mutation mode is not armed",
),
)
]
def worker_checks(targets: Targets) -> list[CheckSpec]:
"""Return the pool, surface, and ephemeral-placeholder checks."""
return [*_pool(targets), *_surfaces(targets), *_ephemeral()]