Coding policy for Viktor Gamov's AI agents: language-agnostic quality rules, autonomous shipping discipline, and stack defaults for JVM, Swift, TypeScript, and Python
93%
Does it follow best practices?
Impact
99%
1.47xAverage score across 10 eval scenarios
Low
Low-risk findings worth noting
#!/usr/bin/env bash
# Outcome-based tests for poll-pr-reviews.sh, focused on the new
# `merge_state` output path: `fetch_merge_state` must return the
# `{status, mergeable}` envelope keyed off GitHub's `mergeStateStatus`
# and `mergeable` fields, and `main` must surface that envelope as a
# top-level field in the snapshot JSON.
#
# Approach: source the script (its main() guard prevents auto-run when
# sourced) and override `gh` with a shell function that returns
# fixture JSON for the two surfaces poll-pr-reviews.sh calls —
# `gh pr checks` and `gh pr view --json mergeStateStatus,mergeable` —
# plus the two `gh api` surfaces for reviews/comments. jq runs locally
# so the script's filter logic is exercised, not duplicated in the test.
#
# Run: bash skills/release/tests/test_poll_pr_reviews.sh
# Exit 0 on all-pass; non-zero with a per-test diagnostic on failure.
set -uo pipefail
# The runner's per-suite token binds this suite's summary line to this
# suite (scripts/run-tests.sh header). Unset it so a harness run from
# here prints an empty token and never stands in for this one.
SUMMARY_TOKEN="${TEST_SUMMARY_TOKEN:-}"
unset TEST_SUMMARY_TOKEN
# The github.com host pin: every gh call this suite's mock sees passes the
# gate first, with GH_HOST pointing at another server (gh_pin_gate.sh).
# shellcheck source=skills/release/tests/gh_pin_gate.sh
source "$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)/gh_pin_gate.sh" \
|| { echo "fatal: cannot source gh_pin_gate.sh" >&2; exit 2; }
# gh_pin_gate.sh turns on -e; this harness counts failures itself and runs
# without it (rules/error-handling.md aggregate-reporting carve-out).
set +e
SCRIPT="$(cd "$(dirname "$0")/.." && pwd)/poll-pr-reviews.sh"
[[ -x "$SCRIPT" ]] || { echo "fatal: poll-pr-reviews.sh not executable at $SCRIPT" >&2; exit 2; }
# shellcheck disable=SC1090
source "$SCRIPT" || true
set +e
FAIL_COUNT=0
PASS_COUNT=0
# The head SHA the mock's `gh pr view` returns. A review fixture whose
# commit_id equals this is "fresh" (bound to head); anything else is "stale"
# and its verdict resolves to state "none" (#186).
HEAD_SHA="head000000000000000000000000000000000000"
OLD_SHA="0ld00000000000000000000000000000000000000"
assert_eq() {
local label="$1" expected="$2" actual="$3"
if [[ "$expected" == "$actual" ]]; then
return 0
fi
echo " FAIL: ${label}: expected '${expected}', got '${actual}'" >&2
return 1
}
run() {
local name="$1"; shift
if "$@"; then
PASS_COUNT=$((PASS_COUNT + 1))
echo " pass: $name"
else
FAIL_COUNT=$((FAIL_COUNT + 1))
echo " FAIL: $name" >&2
fi
}
# Mock `gh` — supports the four invocations poll-pr-reviews.sh issues.
# MOCK_MERGE_STATE selects the fixture for `gh pr view`; other surfaces
# return minimal fixtures so main() can complete.
_gh_mock() {
case "$1" in
pr)
local subcmd="$2"
shift 2
case "$subcmd" in
view)
# Contract: `gh pr view <N> --repo <o/r> --json mergeStateStatus,mergeable`.
# Validate the --json args explicitly so a regression that drops --json
# or asks for the wrong fields surfaces as a loud mock failure rather
# than passing silently against a permissive stub.
local saw_json=0 json_args=""
while [[ $# -gt 0 ]]; do
case "$1" in
--json) saw_json=1; json_args="${2:-}"; shift 2 ;;
*) shift ;;
esac
done
[[ $saw_json -eq 1 ]] || { echo "mock gh pr view: missing --json flag (contract: --json mergeStateStatus,mergeable,headRefOid)" >&2; return 99; }
[[ "$json_args" == "mergeStateStatus,mergeable,headRefOid" ]] || { echo "mock gh pr view: wrong --json args: '${json_args}' (expected 'mergeStateStatus,mergeable,headRefOid')" >&2; return 99; }
# headRefOid is the head SHA review verdicts are bound to. Tests that
# exercise verdict resolution set review fixtures' commit_id to
# HEAD_SHA (fresh) or something else (stale) — see HEAD_SHA below.
case "${MOCK_MERGE_STATE:-}" in
clean) echo '{"mergeStateStatus":"CLEAN","mergeable":"MERGEABLE","headRefOid":"'"$HEAD_SHA"'"}' ;;
dirty) echo '{"mergeStateStatus":"DIRTY","mergeable":"CONFLICTING","headRefOid":"'"$HEAD_SHA"'"}' ;;
unknown) echo '{"mergeStateStatus":"UNKNOWN","mergeable":"UNKNOWN","headRefOid":"'"$HEAD_SHA"'"}' ;;
no-head) echo '{"mergeStateStatus":"CLEAN","mergeable":"MERGEABLE","headRefOid":""}' ;;
*) echo "mock gh: unknown MOCK_MERGE_STATE='${MOCK_MERGE_STATE:-}'" >&2; return 2 ;;
esac
;;
checks)
# gh pr checks <N> --repo <o/r> --json name,bucket
echo "${MOCK_CHECKS_BODY:-[]}"
;;
*) echo "mock gh pr: unsupported subcommand: $subcmd" >&2; return 2 ;;
esac
;;
api)
# `gh api graphql -f query=... --jq ...` — the pending review requests.
# GraphQL, not REST: the REST endpoint omits bot reviewers entirely
# (#276), so every bot lane would read "never requested" there (#369).
#
# Contract-checked, then answered the way GitHub answers: the query must
# select the fields the filter reads and carry the PR as an `Int!`
# variable, and the production `--jq` filter runs over a raw GraphQL
# response. A wrong field selection therefore reads every lane
# unrequested here too, instead of passing against a pre-shaped array.
# MOCK_REQUESTED_BODY lists the pending logins; MOCK_REQUESTED_RAW
# replaces the whole raw response.
if [[ "${2:-}" == "graphql" ]]; then
shift 2
local query="" filter="" pr_var="" owner_var="" repo_var=""
while [[ $# -gt 0 ]]; do
case "$1" in
-f) case "${2:-}" in
query=*) query="${2#query=}" ;;
owner=*) owner_var="${2#owner=}" ;;
repo=*) repo_var="${2#repo=}" ;;
esac; shift 2 ;;
-F) case "${2:-}" in pr=*) pr_var="${2#pr=}" ;; esac; shift 2 ;;
--jq) filter="${2:-}"; shift 2 ;;
*) shift ;;
esac
done
local needle
# shellcheck disable=SC2016 # literal GraphQL variable names the query must carry
for needle in 'reviewRequests(first: 50)' 'requestedReviewer' '... on Bot { login }' \
'... on User { login }' '... on Team { slug }' \
'$owner: String!' '$repo: String!' '$pr: Int!' 'pullRequest(number: $pr)'; do
[[ "$query" == *"$needle"* ]] || { echo "mock gh api graphql: query lacks '${needle}'" >&2; return 99; }
done
[[ "$pr_var" =~ ^[1-9][0-9]*$ ]] || { echo "mock gh api graphql: -F pr must be a positive integer, got '${pr_var}'" >&2; return 99; }
[[ -n "$owner_var" && -n "$repo_var" ]] || { echo "mock gh api graphql: owner and repo must travel as -f variables" >&2; return 99; }
[[ -n "$filter" ]] || { echo "mock gh api graphql: missing --jq" >&2; return 99; }
local raw
raw="${MOCK_REQUESTED_RAW:-$(printf '%s' "${MOCK_REQUESTED_BODY:-[]}" | jq -c \
'{data: {repository: {pullRequest: {reviewRequests: {nodes:
[.[] | {requestedReviewer: {__typename: "Bot", login: .}}]}}}}}')}"
printf '%s' "$raw" | jq "$filter"
return $?
fi
# gh api --paginate repos/<o>/<r>/pulls/<N>/reviews?per_page=100
# gh api --paginate repos/<o>/<r>/pulls/<N>/comments?per_page=100
# The script pipes the raw paginated output through `jq -s` itself,
# so this mock no longer forwards `--jq`. It echoes a fixture body
# keyed off the path; tests can simulate multiple pages by setting
# MOCK_REVIEWS_BODY / MOCK_COMMENTS_BODY to several concatenated
# JSON arrays (what `gh api --paginate` actually emits across pages).
shift # consume "api"
local path="" saw_paginate=0
while [[ $# -gt 0 ]]; do
case "$1" in
--paginate) saw_paginate=1; shift ;;
--jq) echo "mock gh api: --jq is incompatible with --paginate here; the script should jq -s externally" >&2; return 99 ;;
*) [[ -z "$path" ]] && path="$1"; shift ;;
esac
done
[[ $saw_paginate -eq 1 ]] || { echo "mock gh api: missing --paginate (required so the script never silently misses page 2+)" >&2; return 99; }
case "$path" in
*reviews*) echo "${MOCK_REVIEWS_BODY:-[]}" ;;
*comments*) echo "${MOCK_COMMENTS_BODY:-[]}" ;;
*) echo "mock gh api: unsupported path: $path" >&2; return 2 ;;
esac
;;
*) echo "mock gh: unsupported invocation: $*" >&2; return 2 ;;
esac
}
gh() { gh_pin_gate _gh_mock "$@"; }
# --- test bodies ---
# --- pending review requests (#369) ---
# A lane nobody requested is not a lane that is running late: a developer
# reading its own current-tip evidence must be able to tell the two apart
# without sitting out a pre-merge budget on a review it cannot itself request.
t_requested_among_matches_a_pending_bot() {
local out
out=$(requested_among '["github-actions"]' "github-actions[bot]")
assert_eq "requested" "true" "$out"
}
t_requested_among_is_false_when_nobody_asked() {
local out
out=$(requested_among '[]' "github-actions[bot]")
assert_eq "requested" "false" "$out"
}
# The raw GraphQL response, through the production filter: bot suffix and case
# normalized, a team read by slug, a reviewer GitHub could not resolve dropped.
t_fetch_requested_logins_reads_the_raw_graphql_shape() {
local out
MOCK_REQUESTED_RAW='{"data":{"repository":{"pullRequest":{"reviewRequests":{"nodes":[
{"requestedReviewer":{"__typename":"Bot","login":"GitHub-Actions[bot]"}},
{"requestedReviewer":{"__typename":"User","login":"Some-Human"}},
{"requestedReviewer":{"__typename":"Team","slug":"Core-Team"}},
{"requestedReviewer":null}]}}}}}'
out=$(fetch_requested_logins owner repo 7)
unset MOCK_REQUESTED_RAW
assert_eq "logins" '["github-actions","some-human","core-team"]' "$out"
}
# A PR number is validated before any call: `-F` would read a file for an
# `@`-prefixed value, and anything else non-numeric is not a PR.
t_non_integer_pr_number_is_refused_before_any_call() {
local candidate rc out
for candidate in "1) { id } } #" "@/etc/hosts" "0" "-1" "1.5" "" "abc"; do
rc=0
out=$(fetch_requested_logins owner repo "$candidate" 2>&1) || rc=$?
[[ "$rc" -eq 1 && "$out" == *"positive integer"* ]] \
|| { echo " FAIL: fetch_requested_logins accepted '${candidate}' (rc=${rc}): ${out}" >&2; return 1; }
rc=0
out=$( (MOCK_MERGE_STATE=boom main owner repo "$candidate") 2>&1 ) || rc=$?
[[ "$rc" -eq 2 && "$out" == *"positive integer"* ]] \
|| { echo " FAIL: main accepted '${candidate}' (rc=${rc}): ${out}" >&2; return 1; }
done
}
t_requested_among_ignores_another_reviewer() {
local out
out=$(requested_among '["some-human"]' "github-actions[bot]")
assert_eq "requested" "false" "$out"
}
t_main_marks_an_unrequested_lane() {
MOCK_MERGE_STATE=clean
MOCK_REQUESTED_BODY='[]'
local out
out=$(main owner repo 1)
assert_eq "codex state" "none" "$(echo "$out" | jq -r '.reviews.codex.state')" || return 1
assert_eq "codex requested" "false" "$(echo "$out" | jq -r '.reviews.codex.requested')" || return 1
assert_eq "codex stale" "false" "$(echo "$out" | jq -r '.reviews.codex.stale')"
}
t_main_marks_a_pending_lane_requested() {
MOCK_MERGE_STATE=clean
MOCK_REQUESTED_BODY='["github-actions"]'
local out
out=$(main owner repo 1)
assert_eq "codex state" "none" "$(echo "$out" | jq -r '.reviews.codex.state')" || return 1
assert_eq "codex requested" "true" "$(echo "$out" | jq -r '.reviews.codex.requested')"
}
t_fetch_merge_state_clean_returns_mergeable_envelope() {
MOCK_MERGE_STATE=clean
local out status mergeable
out=$(fetch_merge_state "owner" "repo" "1")
status=$(echo "$out" | jq -r '.status')
mergeable=$(echo "$out" | jq -r '.mergeable')
assert_eq "status" "CLEAN" "$status" || return 1
assert_eq "mergeable" "MERGEABLE" "$mergeable"
}
t_fetch_merge_state_dirty_returns_conflicting_envelope() {
MOCK_MERGE_STATE=dirty
local out status mergeable
out=$(fetch_merge_state "owner" "repo" "1")
status=$(echo "$out" | jq -r '.status')
mergeable=$(echo "$out" | jq -r '.mergeable')
assert_eq "status" "DIRTY" "$status" || return 1
assert_eq "mergeable" "CONFLICTING" "$mergeable"
}
t_fetch_merge_state_unknown_returns_unknown_envelope() {
MOCK_MERGE_STATE=unknown
local out status mergeable
out=$(fetch_merge_state "owner" "repo" "1")
status=$(echo "$out" | jq -r '.status')
mergeable=$(echo "$out" | jq -r '.mergeable')
assert_eq "status" "UNKNOWN" "$status" || return 1
assert_eq "mergeable" "UNKNOWN" "$mergeable"
}
t_main_surfaces_merge_state_as_top_level_field() {
MOCK_MERGE_STATE=clean
local out keys
out=$(main "owner" "repo" "1")
keys=$(echo "$out" | jq -r '.merge_state | "\(.status)|\(.mergeable)"')
assert_eq "merge_state in main output" "CLEAN|MERGEABLE" "$keys"
}
t_main_propagates_dirty_state() {
MOCK_MERGE_STATE=dirty
local out keys
out=$(main "owner" "repo" "1")
keys=$(echo "$out" | jq -r '.merge_state | "\(.status)|\(.mergeable)"')
assert_eq "merge_state in main output" "DIRTY|CONFLICTING" "$keys"
}
# Issue #83: on PRs with > 1 page of reviews, gh api without --paginate
# returns only page 1. The pre-fix `| last` filter then picked the last
# entry on page 1 — not the actual newest review on the last page — and
# the gate could approve a merge against stale data.
#
# Build a fixture that mimics what `gh api --paginate` actually emits:
# two concatenated JSON arrays. Page 1's last entry is a COMMENTED review
# at 17:00; page 2's last entry is a CHANGES_REQUESTED review at 18:04.
# A correct implementation must report CHANGES_REQUESTED@18:04.
t_latest_review_by_picks_from_last_page() {
MOCK_REVIEWS_BODY='[{"user":{"login":"github-actions[bot]"},"state":"APPROVED","submitted_at":"2026-05-18T16:00:00Z"},{"user":{"login":"github-actions[bot]"},"state":"COMMENTED","submitted_at":"2026-05-18T17:00:00Z"}][{"user":{"login":"github-actions[bot]"},"state":"CHANGES_REQUESTED","submitted_at":"2026-05-18T18:04:00Z"}]'
local out state submitted_at
out=$(latest_review_by "owner" "repo" "1" "github-actions[bot]")
state=$(echo "$out" | jq -r '.state')
submitted_at=$(echo "$out" | jq -r '.submitted_at')
assert_eq "state from last page" "CHANGES_REQUESTED" "$state" || return 1
assert_eq "submitted_at from last page" "2026-05-18T18:04:00Z" "$submitted_at"
}
t_latest_review_by_returns_none_when_no_reviews() {
MOCK_REVIEWS_BODY='[]'
local out state submitted_at body
out=$(latest_review_by "owner" "repo" "1" "github-actions[bot]")
state=$(echo "$out" | jq -r '.state')
submitted_at=$(echo "$out" | jq -r '.submitted_at')
body=$(echo "$out" | jq -r '.body')
assert_eq "state for empty" "none" "$state" || return 1
assert_eq "submitted_at for empty" "null" "$submitted_at" || return 1
assert_eq "body for empty" "null" "$body"
}
# A review's state classifies merge-gating, not whether its body must be read.
# A COMMENTED review with zero inline comments still carries a body the gate
# must surface so the agent reads it (rules/reviewer-feedback-reading.md).
t_latest_review_by_surfaces_body_text() {
MOCK_REVIEWS_BODY='[{"user":{"login":"github-actions[bot]"},"state":"COMMENTED","submitted_at":"2026-05-18T16:00:00Z","body":"Non-blocking, but the rename in foo.py:42 drops the retry guard."}]'
local out state body
out=$(latest_review_by "owner" "repo" "1" "github-actions[bot]")
state=$(echo "$out" | jq -r '.state')
body=$(echo "$out" | jq -r '.body')
assert_eq "state surfaced" "COMMENTED" "$state" || return 1
assert_eq "body surfaced verbatim" "Non-blocking, but the rename in foo.py:42 drops the retry guard." "$body"
}
t_latest_review_by_filters_other_logins_across_pages() {
# Page 1: two human reviews + one bot review. Page 2: one human review
# that's newer than the bot review. The bot's latest is still the page-1
# bot review, even though the page-2 human is newer.
MOCK_REVIEWS_BODY='[{"user":{"login":"alice"},"state":"COMMENTED","submitted_at":"2026-05-18T15:00:00Z"},{"user":{"login":"github-actions[bot]"},"state":"APPROVED","submitted_at":"2026-05-18T16:00:00Z"},{"user":{"login":"bob"},"state":"COMMENTED","submitted_at":"2026-05-18T16:30:00Z"}][{"user":{"login":"alice"},"state":"COMMENTED","submitted_at":"2026-05-18T17:00:00Z"}]'
local out state submitted_at
out=$(latest_review_by "owner" "repo" "1" "github-actions[bot]")
state=$(echo "$out" | jq -r '.state')
submitted_at=$(echo "$out" | jq -r '.submitted_at')
assert_eq "bot state" "APPROVED" "$state" || return 1
assert_eq "bot submitted_at" "2026-05-18T16:00:00Z" "$submitted_at"
}
# Same shape for comments: counts must sum across pages, not pick page 1
# alone. Mix in a non-target login and an in_reply_to_id to confirm the
# filter still discards both.
t_toplevel_comments_by_sums_across_pages() {
MOCK_COMMENTS_BODY='[{"user":{"login":"github-actions[bot]"},"in_reply_to_id":null},{"user":{"login":"github-actions[bot]"},"in_reply_to_id":null},{"user":{"login":"alice"},"in_reply_to_id":null}][{"user":{"login":"github-actions[bot]"},"in_reply_to_id":null},{"user":{"login":"github-actions[bot]"},"in_reply_to_id":12345}]'
local count
count=$(toplevel_comments_by "owner" "repo" "1" "github-actions[bot]")
assert_eq "top-level bot comments across both pages" "3" "$count"
}
t_toplevel_comments_by_returns_zero_for_no_comments() {
MOCK_COMMENTS_BODY='[]'
local count
count=$(toplevel_comments_by "owner" "repo" "1" "github-actions[bot]")
assert_eq "comments count for empty" "0" "$count"
}
t_toplevel_comments_by_excludes_replies_and_other_logins() {
MOCK_COMMENTS_BODY='[{"user":{"login":"github-actions[bot]"},"in_reply_to_id":null},{"user":{"login":"github-actions[bot]"},"in_reply_to_id":991},{"user":{"login":"some-human"},"in_reply_to_id":null}]'
local count
count=$(toplevel_comments_by "owner" "repo" "1" "${CODEX_COMMENT_LOGINS[@]}")
assert_eq "replies and foreign logins excluded" "1" "$count"
}
# A degraded API response can put a non-object element (an error body, a bare
# string) in the array. Indexing it aborted jq mid-filter and failed the whole
# poll: watch-pr-reviews.sh then burned all 50 attempts and reported
# `pending_at_budget`, which reads as "no reviewer posted" rather than "the
# tool broke" (issue #300). Real comments around it must still be counted.
t_toplevel_comments_by_survives_non_object_element() {
MOCK_COMMENTS_BODY='["Not Found",{"user":{"login":"github-actions[bot]"},"in_reply_to_id":null},{"user":{"login":"github-actions[bot]"},"in_reply_to_id":null}]'
local count
count=$(toplevel_comments_by "owner" "repo" "1" "${CODEX_COMMENT_LOGINS[@]}" 2>/dev/null)
assert_eq "count survives a non-object element" "2" "$count"
}
# The drop is visible, never silent: best-effort work that continues past a
# failure warns (rules/error-handling.md Shell Error Handling).
t_toplevel_comments_by_warns_on_non_object_element() {
MOCK_COMMENTS_BODY='["Not Found",{"user":{"login":"github-actions[bot]"},"in_reply_to_id":null}]'
local err
err=$( { toplevel_comments_by "owner" "repo" "1" "${CODEX_COMMENT_LOGINS[@]}" >/dev/null; } 2>&1 )
case "$err" in
*"1 non-object element"*) gh_hints_pinned "$err" ;;
*) echo " FAIL: expected a non-object-element warning on stderr, got: '${err}'" >&2; return 1 ;;
esac
}
# Same guard on the reviews surface — the merge gate reads it, and a
# non-object element carries no verdict, so dropping it cannot mask one.
t_latest_review_by_survives_non_object_element() {
MOCK_REVIEWS_BODY='["Not Found",{"user":{"login":"github-actions[bot]"},"state":"APPROVED","submitted_at":"2026-05-18T16:00:00Z","body":"ok","commit_id":"'"$HEAD_SHA"'"}]'
local out state
out=$(latest_review_by "owner" "repo" "1" "github-actions[bot]" 2>/dev/null)
state=$(echo "$out" | jq -r '.state')
assert_eq "verdict survives a non-object element" "APPROVED" "$state"
}
# End-to-end: the poll returns a snapshot instead of failing, which is what
# kept watch-pr-reviews.sh from stalling to budget on a degraded response.
t_main_survives_non_object_comment_element() {
MOCK_MERGE_STATE=clean
MOCK_COMMENTS_BODY='["Not Found",{"user":{"login":"github-actions[bot]"},"in_reply_to_id":null}]'
local out codex
out=$(main "owner" "repo" "1" 2>/dev/null) || return 1
codex=$(echo "$out" | jq -r '.inline_comments.codex')
assert_eq "snapshot still produced, Codex comment counted" "1" "$codex"
}
# End-to-end through main() — the path the Step 7 merge gate actually reads.
# Copilot has no lane (docs/adr/0014-retire-copilot-lane.md): its comments
# never count as Codex's, and land in `other` so each still needs a reply.
t_main_counts_copilot_comments_as_other() {
MOCK_MERGE_STATE=clean
MOCK_COMMENTS_BODY='[{"user":{"login":"Copilot"},"in_reply_to_id":null},{"user":{"login":"github-actions[bot]"},"in_reply_to_id":null},{"user":{"login":"github-actions[bot]"},"in_reply_to_id":null}]'
local out counts
out=$(main "owner" "repo" "1")
counts=$(echo "$out" | jq -c '.inline_comments')
assert_eq "inline_comments" '{"codex":2,"other":1}' "$counts"
}
# The resolver is variadic on login: a second policy-reviewer identity (a
# GitHub App, say) is resolved by passing it explicitly. These cases supply a
# hypothetical `policy-app[bot]` alongside the shipped login to prove the
# multi-login aggregation, without shipping that login in CODEX_REVIEW_LOGINS.
t_latest_review_by_resolves_extra_login() {
MOCK_REVIEWS_BODY='[{"user":{"login":"policy-app[bot]"},"state":"APPROVED","submitted_at":"2026-07-21T05:15:00Z"}]'
local out state
out=$(latest_review_by "owner" "repo" "1" "${CODEX_REVIEW_LOGINS[@]}" "policy-app[bot]")
state=$(echo "$out" | jq -r '.state')
assert_eq "extra-login review resolves to policy state" "APPROVED" "$state"
}
# Resolving across two logins must return the newest verdict when both appear.
t_latest_review_by_policy_reviewer_picks_newest_across_logins() {
MOCK_REVIEWS_BODY='[{"user":{"login":"github-actions[bot]"},"state":"COMMENTED","submitted_at":"2026-07-21T05:00:00Z"},{"user":{"login":"policy-app[bot]"},"state":"CHANGES_REQUESTED","submitted_at":"2026-07-21T05:10:00Z"}]'
local out state
out=$(latest_review_by "owner" "repo" "1" "${CODEX_REVIEW_LOGINS[@]}" "policy-app[bot]")
state=$(echo "$out" | jq -r '.state')
assert_eq "newest policy verdict across both logins" "CHANGES_REQUESTED" "$state"
}
# Fail-safe: an active CHANGES_REQUESTED from one policy-reviewer identity must
# NOT be masked by a LATER clean review from the other identity — a gating
# watcher that surfaced the newest-overall would think the reviewer was clean
# while GitHub still blocked the merge.
t_latest_review_by_changes_requested_not_masked_by_later_other_login() {
MOCK_REVIEWS_BODY='[{"user":{"login":"github-actions[bot]"},"state":"CHANGES_REQUESTED","submitted_at":"2026-07-21T05:00:00Z"},{"user":{"login":"policy-app[bot]"},"state":"COMMENTED","submitted_at":"2026-07-21T05:10:00Z"}]'
local out state
out=$(latest_review_by "owner" "repo" "1" "${CODEX_REVIEW_LOGINS[@]}" "policy-app[bot]")
state=$(echo "$out" | jq -r '.state')
assert_eq "active block not masked by later clean review" "CHANGES_REQUESTED" "$state"
}
# An unknown login is NOT the policy reviewer: its review must not surface as
# .reviews.codex, and its comments must not count toward the reply gate.
t_main_ignores_unknown_login_review() {
MOCK_MERGE_STATE=clean
MOCK_REVIEWS_BODY='[{"user":{"login":"some-other-app[bot]"},"state":"APPROVED","submitted_at":"2026-07-21T05:15:00Z","commit_id":"'"$HEAD_SHA"'"}]'
local out state
out=$(main "owner" "repo" "1")
state=$(echo "$out" | jq -r '.reviews.codex.state')
assert_eq "unknown-login review is not the policy reviewer" "none" "$state"
}
t_toplevel_comments_by_ignores_unknown_login() {
MOCK_COMMENTS_BODY='[{"user":{"login":"some-other-app[bot]"},"in_reply_to_id":null}]'
local count
count=$(toplevel_comments_by "owner" "repo" "1" "${CODEX_COMMENT_LOGINS[@]}")
assert_eq "unknown-login inline comment not counted for policy reviewer" "0" "$count"
}
# A login's own latest review must be selected by submitted_at, NOT by the
# API's array position: jq group_by preserves input order within a group, so
# if the reviews API returns a login's reviews out of chronological order, a
# `last`-based pick would gate on a superseded verdict. Here the newest review
# (CHANGES_REQUESTED @18:00) is NOT last in the input — a correct max_by pick
# surfaces it; a `last` pick would wrongly return the APPROVED @16:00.
t_latest_review_by_picks_max_by_time_not_array_position() {
MOCK_REVIEWS_BODY='[{"user":{"login":"github-actions[bot]"},"state":"CHANGES_REQUESTED","submitted_at":"2026-07-25T18:00:00Z"},{"user":{"login":"github-actions[bot]"},"state":"COMMENTED","submitted_at":"2026-07-25T17:00:00Z"},{"user":{"login":"github-actions[bot]"},"state":"APPROVED","submitted_at":"2026-07-25T16:00:00Z"}]'
local out state submitted_at
out=$(latest_review_by "owner" "repo" "1" "github-actions[bot]")
state=$(echo "$out" | jq -r '.state')
submitted_at=$(echo "$out" | jq -r '.submitted_at')
assert_eq "latest by time, not array position" "CHANGES_REQUESTED" "$state" || return 1
assert_eq "latest submitted_at" "2026-07-25T18:00:00Z" "$submitted_at"
}
# GitHub review states outside the snapshot schema (DISMISSED, PENDING) are
# not live verdicts, but watch-pr-reviews.sh treats any non-"none" state as
# "a bot posted" and would let one satisfy the ready gate. They must normalize
# to "none".
t_latest_review_by_normalizes_dismissed_to_none() {
MOCK_REVIEWS_BODY='[{"user":{"login":"github-actions[bot]"},"state":"DISMISSED","submitted_at":"2026-07-25T16:00:00Z","body":"was dismissed"}]'
local out state
out=$(latest_review_by "owner" "repo" "1" "github-actions[bot]")
state=$(echo "$out" | jq -r '.state')
assert_eq "DISMISSED normalizes to none" "none" "$state"
}
t_latest_review_by_normalizes_pending_to_none() {
MOCK_REVIEWS_BODY='[{"user":{"login":"github-actions[bot]"},"state":"PENDING","submitted_at":"2026-07-25T16:00:00Z"}]'
local out state
out=$(latest_review_by "owner" "repo" "1" "github-actions[bot]")
state=$(echo "$out" | jq -r '.state')
assert_eq "PENDING normalizes to none" "none" "$state"
}
# --- #186: review verdicts bound to the PR head SHA ---
# A verdict on the current head passes through unchanged, flagged not-stale.
t_resolve_against_head_fresh_passes_through() {
local review out state stale
review='{"state":"APPROVED","submitted_at":"2026-07-25T10:00:00Z","body":"ok","commit_id":"'"$HEAD_SHA"'"}'
out=$(resolve_review_against_head "$review" "$HEAD_SHA")
state=$(echo "$out" | jq -r '.state')
stale=$(echo "$out" | jq -r '.stale')
assert_eq "fresh review keeps its state" "APPROVED" "$state" || return 1
assert_eq "fresh review not stale" "false" "$stale"
}
# A verdict on a superseded SHA collapses to "none" (absent, not clean) with
# stale=true, but keeps its body/commit_id visible for diagnosis. This is the
# #186 fix: a stale APPROVED/COMMENTED must not read as a live clean verdict.
t_resolve_against_head_stale_collapses_to_none() {
local review out state stale body commit
review='{"state":"APPROVED","submitted_at":"2026-07-25T09:00:00Z","body":"reviewed old code","commit_id":"'"$OLD_SHA"'"}'
out=$(resolve_review_against_head "$review" "$HEAD_SHA")
state=$(echo "$out" | jq -r '.state')
stale=$(echo "$out" | jq -r '.stale')
body=$(echo "$out" | jq -r '.body')
commit=$(echo "$out" | jq -r '.commit_id')
assert_eq "stale review resolves to none" "none" "$state" || return 1
assert_eq "stale review flagged stale" "true" "$stale" || return 1
assert_eq "stale verdict body kept" "reviewed old code" "$body" || return 1
assert_eq "stale verdict commit kept" "$OLD_SHA" "$commit"
}
# A genuinely-absent review (never posted) stays none, stale=false — an agent
# can tell "never reviewed" (stale false) from "reviewed older commit" (true).
t_resolve_against_head_none_stays_none_not_stale() {
local review out state stale
review='{"state":"none","submitted_at":null,"body":null,"commit_id":null}'
out=$(resolve_review_against_head "$review" "$HEAD_SHA")
state=$(echo "$out" | jq -r '.state')
stale=$(echo "$out" | jq -r '.stale')
assert_eq "absent review stays none" "none" "$state" || return 1
assert_eq "absent review is not stale" "false" "$stale"
}
# End-to-end #186 symptom 2 (false ready): a stale COMMENTED from an earlier
# SHA must NOT surface as a live verdict — it collapses to none so the watcher
# keeps waiting instead of merging unreviewed code.
t_main_stale_review_collapses_to_none() {
MOCK_MERGE_STATE=clean
MOCK_REVIEWS_BODY='[{"user":{"login":"github-actions[bot]"},"state":"COMMENTED","submitted_at":"2026-07-25T09:00:00Z","body":"looks fine","commit_id":"'"$OLD_SHA"'"}]'
local out state stale
out=$(main "owner" "repo" "1")
state=$(echo "$out" | jq -r '.reviews.codex.state')
stale=$(echo "$out" | jq -r '.reviews.codex.stale')
assert_eq "stale codex verdict reads none in snapshot" "none" "$state" || return 1
assert_eq "stale codex verdict flagged stale" "true" "$stale"
}
# A fresh verdict (commit_id == head) surfaces normally.
t_main_fresh_review_surfaces_state() {
MOCK_MERGE_STATE=clean
MOCK_REVIEWS_BODY='[{"user":{"login":"github-actions[bot]"},"state":"APPROVED","submitted_at":"2026-07-25T10:00:00Z","body":"ok","commit_id":"'"$HEAD_SHA"'"}]'
local out state stale
out=$(main "owner" "repo" "1")
state=$(echo "$out" | jq -r '.reviews.codex.state')
stale=$(echo "$out" | jq -r '.reviews.codex.stale')
assert_eq "fresh codex verdict surfaces" "APPROVED" "$state" || return 1
assert_eq "fresh codex verdict not stale" "false" "$stale"
}
t_main_surfaces_head_sha_top_level() {
MOCK_MERGE_STATE=clean
MOCK_REVIEWS_BODY='[]'
local out head
out=$(main "owner" "repo" "1")
head=$(echo "$out" | jq -r '.head_sha')
assert_eq "head_sha surfaced top-level" "$HEAD_SHA" "$head"
}
# Guard: an empty headRefOid must fail loudly, never silently void the review
# gate by marking every verdict stale.
t_main_no_head_sha_fails() {
MOCK_MERGE_STATE=no-head
MOCK_REVIEWS_BODY='[]'
# Subshell: main() calls `exit` on the guard, which would otherwise
# terminate this sourced test script rather than just the call.
local rc=0 err
err=$( ( main "owner" "repo" "1" >/dev/null ) 2>&1 ) || rc=$?
[[ "$rc" -ne 0 ]] || { echo " FAIL: main should exit non-zero on empty headRefOid, got rc=0" >&2; return 1; }
gh_hints_pinned "$err"
}
# --- #182: cancelled superseded runs are no-signal, not failure ---
# A cancelled bucket alongside a success is still success (the cancelled twin
# was superseded; its replacement concluded). Folding cancel into pending would
# have wedged this.
t_ci_status_cancel_with_success_is_success() {
MOCK_MERGE_STATE=clean
MOCK_CHECKS_BODY='[{"name":"CI","bucket":"pass"},{"name":"CI","bucket":"cancel"}]'
local out ci
out=$(main "owner" "repo" "1")
ci=$(echo "$out" | jq -r '.ci.status')
assert_eq "success alongside a cancelled twin is success" "success" "$ci"
}
# Only cancels present (replacement not yet registered) → pending, so the
# watcher waits rather than concluding.
t_ci_status_only_cancels_is_pending() {
MOCK_MERGE_STATE=clean
MOCK_CHECKS_BODY='[{"name":"CI","bucket":"cancel"},{"name":"review","bucket":"cancel"}]'
local out ci
out=$(main "owner" "repo" "1")
ci=$(echo "$out" | jq -r '.ci.status')
assert_eq "all-cancelled reads pending" "pending" "$ci"
}
# Fixtures below use the exact `gh pr checks --json name,bucket` shape this
# repo's own PRs return. Supersession is per check name: a cancel is settled
# only by a live run of the SAME name on head, never by other checks passing.
# A cancelled required check with no replacement, next to other passing
# checks, stays pending — never success, so the watcher never reports ready.
t_ci_status_unreplaced_cancel_with_other_passes_is_pending() {
MOCK_MERGE_STATE=clean
MOCK_CHECKS_BODY='[{"bucket":"cancel","name":"unit-tests"},{"bucket":"pass","name":"review"},{"bucket":"pass","name":"lint"},{"bucket":"pass","name":"changelog-fragment"},{"bucket":"pass","name":"diagnostics"}]'
local out ci
out=$(main "owner" "repo" "1")
ci=$(echo "$out" | jq -r '.ci.status')
assert_eq "other passing checks do not settle an unreplaced cancel" "pending" "$ci"
}
# The cancelled check re-ran on head and passed: its replacement's result
# drives the state.
t_ci_status_cancel_replaced_by_pass_is_success() {
MOCK_MERGE_STATE=clean
MOCK_CHECKS_BODY='[{"bucket":"cancel","name":"unit-tests"},{"bucket":"pass","name":"unit-tests"},{"bucket":"pass","name":"review"},{"bucket":"pass","name":"lint"}]'
local out ci
out=$(main "owner" "repo" "1")
ci=$(echo "$out" | jq -r '.ci.status')
assert_eq "a passing same-name replacement settles the cancel" "success" "$ci"
}
# The cancelled check re-ran on head and failed: the replacement's failure
# is the result.
t_ci_status_cancel_replaced_by_fail_is_failure() {
MOCK_MERGE_STATE=clean
MOCK_CHECKS_BODY='[{"bucket":"cancel","name":"unit-tests"},{"bucket":"fail","name":"unit-tests"},{"bucket":"pass","name":"review"},{"bucket":"pass","name":"lint"}]'
local out ci
out=$(main "owner" "repo" "1")
ci=$(echo "$out" | jq -r '.ci.status')
assert_eq "a failing same-name replacement is failure" "failure" "$ci"
}
# One cancelled check's replacement cannot settle a different cancelled check.
t_ci_status_replacement_does_not_cover_another_cancel() {
MOCK_MERGE_STATE=clean
MOCK_CHECKS_BODY='[{"bucket":"cancel","name":"unit-tests"},{"bucket":"pass","name":"unit-tests"},{"bucket":"cancel","name":"diagnostics"},{"bucket":"pass","name":"lint"}]'
local out ci
out=$(main "owner" "repo" "1")
ci=$(echo "$out" | jq -r '.ci.status')
assert_eq "each cancelled check needs its own replacement" "pending" "$ci"
}
# A real failure on head still fails, even next to a cancelled run.
t_ci_status_fail_with_cancel_is_failure() {
MOCK_MERGE_STATE=clean
MOCK_CHECKS_BODY='[{"name":"CI","bucket":"fail"},{"name":"old","bucket":"cancel"}]'
local out ci
out=$(main "owner" "repo" "1")
ci=$(echo "$out" | jq -r '.ci.status')
assert_eq "a real fail still fails next to a cancel" "failure" "$ci"
}
# --- review lanes (docs/adr/0010-review-lanes.md) ---
# Codex and Claude both post as github-actions[bot]; reviews.codex must read
# the codex-policy lane only, and the lanes view must report every lane.
t_main_reads_codex_verdict_per_lane() {
MOCK_MERGE_STATE=clean
MOCK_REVIEWS_BODY='[
{"id":1,"user":{"login":"github-actions[bot]","type":"Bot"},"state":"CHANGES_REQUESTED","submitted_at":"2026-10-20T10:00:00Z","commit_id":"'"$HEAD_SHA"'","body":"<!-- coding-policy:codex-policy-review:v1 -->\nPolicy loaded: 30 rule files. One violation."},
{"id":2,"user":{"login":"github-actions[bot]","type":"Bot"},"state":"COMMENTED","submitted_at":"2026-10-20T10:05:00Z","commit_id":"'"$HEAD_SHA"'","body":"<!-- coding-policy:claude-quality-review:v1 -->\nNo correctness issues found."},
{"id":4,"user":{"login":"copilot-pull-request-reviewer[bot]","type":"Bot"},"state":"COMMENTED","submitted_at":"2026-10-20T10:10:00Z","commit_id":"'"$HEAD_SHA"'","body":"Copilot reviewed 3 files."}]'
local out
out=$(main owner repo 1)
unset MOCK_REVIEWS_BODY
assert_eq "reviews.codex state" "CHANGES_REQUESTED" "$(jq -r '.reviews.codex.state' <<<"$out")" || return 1
assert_eq "codex lane gating" "true" "$(jq -r '.lanes["codex-policy"].gating' <<<"$out")" || return 1
assert_eq "claude lane state" "COMMENTED" "$(jq -r '.lanes["claude-quality"].state' <<<"$out")" || return 1
assert_eq "claude lane gating" "false" "$(jq -r '.lanes["claude-quality"].gating' <<<"$out")" || return 1
assert_eq "lane set" '["claude-quality","codex-policy"]' "$(jq -c '.lanes | keys' <<<"$out")" || return 1
assert_eq "review keys" '["codex"]' "$(jq -c '.reviews | keys' <<<"$out")" || return 1
assert_eq "copilot is an other review" '[{"login":"copilot-pull-request-reviewer[bot]","is_bot":true,"state":"COMMENTED","body":"Copilot reviewed 3 files."}]' \
"$(jq -c '.other_reviews | map({login, is_bot, state, body})' <<<"$out")"
}
# A human CHANGES_REQUESTED on an older commit stays CHANGES_REQUESTED: GitHub
# keeps it active across pushes, so it is not collapsed against head.
t_main_reports_human_review_unresolved_by_head() {
MOCK_MERGE_STATE=clean
MOCK_REVIEWS_BODY='[
{"id":5,"user":{"login":"alice","type":"User"},"state":"COMMENTED","submitted_at":"2026-10-20T09:00:00Z","commit_id":"'"$OLD_SHA"'","body":"first look"},
{"id":6,"user":{"login":"alice","type":"User"},"state":"CHANGES_REQUESTED","submitted_at":"2026-10-20T09:30:00Z","commit_id":"'"$OLD_SHA"'","body":"please rename"},
{"id":7,"user":{"login":"github-actions[bot]","type":"Bot"},"state":"COMMENTED","submitted_at":"2026-10-20T10:00:00Z","commit_id":"'"$HEAD_SHA"'","body":"<!-- coding-policy:codex-policy-review:v1 -->\nclean"}]'
local out
out=$(main owner repo 1)
unset MOCK_REVIEWS_BODY
assert_eq "human review kept" '[{"login":"alice","is_bot":false,"state":"CHANGES_REQUESTED","body":"please rename","stale":true}]' \
"$(jq -c '.other_reviews | map({login, is_bot, state, body, stale})' <<<"$out")"
}
t_main_unmarked_review_is_codex_lane() {
MOCK_MERGE_STATE=clean
MOCK_REVIEWS_BODY='[{"id":3,"user":{"login":"github-actions[bot]","type":"Bot"},"state":"COMMENTED","submitted_at":"2026-10-20T10:00:00Z","commit_id":"'"$HEAD_SHA"'","body":"Policy loaded: 30 rule files. All rules pass."}]'
local out
out=$(main owner repo 1)
unset MOCK_REVIEWS_BODY
assert_eq "legacy review is codex" "COMMENTED" "$(jq -r '.reviews.codex.state' <<<"$out")" || return 1
assert_eq "claude lane empty" "none" "$(jq -r '.lanes["claude-quality"].state' <<<"$out")"
}
# GitHub resolves a reviewer's state from decisive reviews only: a later
# COMMENTED (a thread reply is one) never clears CHANGES_REQUESTED.
human_state_after() { # <state1> <state2> -> other_reviews[0] {state, body}
MOCK_MERGE_STATE=clean
MOCK_REVIEWS_BODY='[
{"id":5,"user":{"login":"alice","type":"User"},"state":"'"$1"'","submitted_at":"2026-10-20T09:00:00Z","commit_id":"'"$OLD_SHA"'","body":"first"},
{"id":6,"user":{"login":"alice","type":"User"},"state":"'"$2"'","submitted_at":"2026-10-20T09:30:00Z","commit_id":"'"$HEAD_SHA"'","body":"second"}]'
local out
out=$(main owner repo 1)
unset MOCK_REVIEWS_BODY
jq -c '.other_reviews | map({state, body})' <<<"$out"
}
t_human_comment_does_not_clear_changes_requested() {
assert_eq "active verdict" '[{"state":"CHANGES_REQUESTED","body":"first"}]' "$(human_state_after CHANGES_REQUESTED COMMENTED)"
}
# Every body stays readable, not only the one that decided the verdict.
t_human_reviews_keep_every_body() {
MOCK_MERGE_STATE=clean
MOCK_REVIEWS_BODY='[
{"id":6,"user":{"login":"alice","type":"User"},"state":"COMMENTED","submitted_at":"2026-10-20T09:30:00Z","commit_id":"'"$HEAD_SHA"'","body":"second"},
{"id":5,"user":{"login":"alice","type":"User"},"state":"CHANGES_REQUESTED","submitted_at":"2026-10-20T09:00:00Z","commit_id":"'"$OLD_SHA"'","body":"first"}]'
local out
out=$(main owner repo 1)
unset MOCK_REVIEWS_BODY
assert_eq "every body" '[[{"state":"CHANGES_REQUESTED","body":"first"},{"state":"COMMENTED","body":"second"}]]' \
"$(jq -c '.other_reviews | map(.reviews | map({state, body}))' <<<"$out")"
}
t_human_approval_clears_changes_requested() {
assert_eq "active verdict" '[{"state":"APPROVED","body":"second"}]' "$(human_state_after CHANGES_REQUESTED APPROVED)"
}
t_human_dismissal_then_comment_reads_commented() {
assert_eq "active verdict" '[{"state":"COMMENTED","body":"second"}]' "$(human_state_after DISMISSED COMMENTED)"
}
claude_expected_with_checks() { # <checks-json>
MOCK_MERGE_STATE=clean
MOCK_CHECKS_BODY="$1"
local out
out=$(main owner repo 1)
unset MOCK_CHECKS_BODY
jq -c '[.lanes["codex-policy"].expected, .lanes["claude-quality"].expected]' <<<"$out"
}
t_claude_lane_expected_when_workflow_ran() {
assert_eq "expected" '[true,true]' \
"$(claude_expected_with_checks '[{"name":"review","bucket":"pending","workflow":"PR Code Review (Claude, subscription)"}]')"
}
t_claude_lane_not_expected_without_workflow() {
assert_eq "expected" '[true,false]' \
"$(claude_expected_with_checks '[{"name":"review","bucket":"pass","workflow":"PR Policy Review (Codex, subscription)"}]')"
}
t_claude_lane_not_expected_when_skipped() {
assert_eq "expected" '[true,false]' \
"$(claude_expected_with_checks '[{"name":"review","bucket":"skipping","workflow":"PR Code Review (Claude, subscription)"}]')"
}
# --- driver ---
echo "== poll-pr-reviews.sh tests =="
run "fetch_merge_state returns {CLEAN, MERGEABLE} for a clean PR" t_fetch_merge_state_clean_returns_mergeable_envelope
run "fetch_merge_state returns {DIRTY, CONFLICTING} on conflict" t_fetch_merge_state_dirty_returns_conflicting_envelope
run "fetch_merge_state propagates UNKNOWN/UNKNOWN while computing" t_fetch_merge_state_unknown_returns_unknown_envelope
run "main surfaces merge_state as a top-level field" t_main_surfaces_merge_state_as_top_level_field
run "main propagates DIRTY merge_state end-to-end" t_main_propagates_dirty_state
run "latest_review_by picks newest review on page 2 (issue #83)" t_latest_review_by_picks_from_last_page
run "latest_review_by returns 'none' for empty reviews" t_latest_review_by_returns_none_when_no_reviews
run "latest_review_by surfaces the review body text" t_latest_review_by_surfaces_body_text
run "latest_review_by ignores other logins across pages" t_latest_review_by_filters_other_logins_across_pages
run "toplevel_comments_by sums counts across pages (issue #83)" t_toplevel_comments_by_sums_across_pages
run "toplevel_comments_by returns 0 for empty comments" t_toplevel_comments_by_returns_zero_for_no_comments
run "toplevel_comments_by excludes replies and foreign logins" t_toplevel_comments_by_excludes_replies_and_other_logins
run "main counts Copilot comments as other" t_main_counts_copilot_comments_as_other
run "main reports a human review without head collapse" t_main_reports_human_review_unresolved_by_head
run "a later human COMMENTED never clears CHANGES_REQUESTED" t_human_comment_does_not_clear_changes_requested
run "every unlaned review body stays in the snapshot" t_human_reviews_keep_every_body
run "a later human APPROVED clears CHANGES_REQUESTED" t_human_approval_clears_changes_requested
run "a dismissed human block reads as the later comment" t_human_dismissal_then_comment_reads_commented
run "the Claude lane is expected when its workflow ran on head" t_claude_lane_expected_when_workflow_ran
run "the Claude lane is not expected without its workflow" t_claude_lane_not_expected_without_workflow
run "a skipped Claude workflow does not make its lane expected" t_claude_lane_not_expected_when_skipped
run "toplevel_comments_by survives a non-object element (#300)" t_toplevel_comments_by_survives_non_object_element
run "toplevel_comments_by warns about the dropped element (#300)" t_toplevel_comments_by_warns_on_non_object_element
run "latest_review_by survives a non-object element (#300)" t_latest_review_by_survives_non_object_element
run "main returns a snapshot despite a non-object element (#300)" t_main_survives_non_object_comment_element
run "latest_review_by resolves an explicitly passed extra login" t_latest_review_by_resolves_extra_login
run "latest_review_by picks newest policy verdict across logins" t_latest_review_by_policy_reviewer_picks_newest_across_logins
run "latest_review_by never masks an active block with a later clean" t_latest_review_by_changes_requested_not_masked_by_later_other_login
run "latest_review_by picks latest by time, not array position" t_latest_review_by_picks_max_by_time_not_array_position
run "latest_review_by normalizes DISMISSED to none" t_latest_review_by_normalizes_dismissed_to_none
run "latest_review_by normalizes PENDING to none" t_latest_review_by_normalizes_pending_to_none
run "main does not surface an unknown login as .reviews.codex" t_main_ignores_unknown_login_review
run "toplevel_comments_by ignores an unknown comment login" t_toplevel_comments_by_ignores_unknown_login
run "resolve_review_against_head: fresh verdict passes through" t_resolve_against_head_fresh_passes_through
run "resolve_review_against_head: stale verdict collapses to none" t_resolve_against_head_stale_collapses_to_none
run "resolve_review_against_head: absent stays none, not stale" t_resolve_against_head_none_stays_none_not_stale
run "main collapses a stale review to none (#186 false ready)" t_main_stale_review_collapses_to_none
run "main surfaces a fresh (head-bound) review's state" t_main_fresh_review_surfaces_state
run "main surfaces head_sha as a top-level field" t_main_surfaces_head_sha_top_level
run "main fails loudly on an empty headRefOid" t_main_no_head_sha_fails
run "ci.status: success next to a cancelled twin is success (#182)" t_ci_status_cancel_with_success_is_success
run "ci.status: only-cancels reads pending (#182)" t_ci_status_only_cancels_is_pending
run "ci.status: an unreplaced cancel next to passes is pending" t_ci_status_unreplaced_cancel_with_other_passes_is_pending
run "ci.status: a cancel replaced by a pass is success" t_ci_status_cancel_replaced_by_pass_is_success
run "ci.status: a cancel replaced by a fail is failure" t_ci_status_cancel_replaced_by_fail_is_failure
run "ci.status: one replacement cannot settle another cancel" t_ci_status_replacement_does_not_cover_another_cancel
run "ci.status: a real fail next to a cancel still fails (#182)" t_ci_status_fail_with_cancel_is_failure
run "requested_among: a pending bot request matches (#369)" t_requested_among_matches_a_pending_bot
run "requested_among: no pending request reads false (#369)" t_requested_among_is_false_when_nobody_asked
run "requested_among: another reviewer's request is not this one" t_requested_among_ignores_another_reviewer
run "fetch_requested_logins filters the raw GraphQL response" t_fetch_requested_logins_reads_the_raw_graphql_shape
run "a non-integer PR number is refused before any call" t_non_integer_pr_number_is_refused_before_any_call
run "main marks a lane nobody requested (#369)" t_main_marks_an_unrequested_lane
run "main marks a requested-and-pending lane (#369)" t_main_marks_a_pending_lane_requested
run "main reads the Codex verdict from its own lane" t_main_reads_codex_verdict_per_lane
run "main classifies an unmarked legacy review as Codex" t_main_unmarked_review_is_codex_lane
echo "== summary: ${PASS_COUNT} passed, ${FAIL_COUNT} failed =="
echo "test-summary: passed=${PASS_COUNT} failed=${FAIL_COUNT} token=${SUMMARY_TOKEN}"
[[ "$FAIL_COUNT" -eq 0 ]].tessl-plugin
hooks
rules
skills
herdr-standup
herdr-teamlead
references
teamlead
templates
tests
onboard-repo
release
tests