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 review-lanes.sh: the executable's lane config and
# the sourced `review_lane($cfg)` classifier over real-shaped GitHub review
# objects (docs/adr/0010-review-lanes.md).
#
# Run: bash skills/release/tests/test_review_lanes.sh
# Exit 0 on all-pass; non-zero with a per-test diagnostic on failure.
set -uo pipefail
# The runner's per-suite token binds this suite's summary line to this
# suite (scripts/run-tests.sh header). Unset it so a harness run from
# here prints an empty token and never stands in for this one.
SUMMARY_TOKEN="${TEST_SUMMARY_TOKEN:-}"
unset TEST_SUMMARY_TOKEN
SCRIPT="$(cd "$(dirname "$0")/.." && pwd)/review-lanes.sh"
[[ -r "$SCRIPT" ]] || { echo "fatal: review-lanes.sh not readable at $SCRIPT" >&2; exit 2; }
command -v jq >/dev/null 2>&1 || { echo "fatal: jq is required to run these tests" >&2; exit 2; }
# shellcheck source=skills/release/review-lanes.sh
source "$SCRIPT" || { echo "fatal: cannot source $SCRIPT" >&2; exit 2; }
set +e
FAIL_COUNT=0
PASS_COUNT=0
assert_eq() {
local label="$1" expected="$2" actual="$3"
[[ "$expected" == "$actual" ]] && return 0
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: $name" >&2
fi
}
# Classify one review object through the shipped config and jq defs.
lane_of() { # <login> <body-json-string-or-null>
local cfg
cfg=$(bash "$SCRIPT") || return 1
jq -cn --arg login "$1" --argjson body "$2" \
'{id: 1, user: {login: $login, type: "Bot"}, state: "COMMENTED", body: $body}' \
| jq -r --argjson cfg "$cfg" "$(review_lanes_jq) review_lane(\$cfg)"
}
t_config_shape() {
local cfg
cfg=$(bash "$SCRIPT") || return 1
assert_eq "codex gates" "true" "$(jq -r '.["codex-policy"].gating' <<<"$cfg")" || return 1
assert_eq "claude gates" "false" "$(jq -r '.["claude-quality"].gating' <<<"$cfg")" || return 1
assert_eq "lane set" '["claude-quality","codex-policy"]' "$(jq -c 'keys' <<<"$cfg")" || return 1
assert_eq "codex marker" "<!-- coding-policy:codex-policy-review:v1 -->" "$(jq -r '.["codex-policy"].marker' <<<"$cfg")" || return 1
assert_eq "claude marker" "<!-- coding-policy:claude-quality-review:v1 -->" "$(jq -r '.["claude-quality"].marker' <<<"$cfg")" || return 1
assert_eq "codex workflow" "null" "$(jq -r '.["codex-policy"].workflow' <<<"$cfg")" || return 1
assert_eq "claude workflow" "PR Code Review (Claude, subscription)" "$(jq -r '.["claude-quality"].workflow' <<<"$cfg")"
}
# The Claude lane is expected by its workflow's check name; a renamed
# workflow would silently stop the watcher waiting for it.
t_claude_workflow_matches_the_shipped_workflow() {
local wf name
wf="$(cd "$(dirname "$SCRIPT")/../.." && pwd)/.github/workflows/review-claude.yml"
name=$(sed -n 's/^name: //p' "$wf") || return 1
assert_eq "workflow name" "$name" "$(bash "$SCRIPT" | jq -r '.["claude-quality"].workflow')"
}
t_claude_marker_under_shared_login() {
assert_eq "lane" "claude-quality" "$(lane_of 'github-actions[bot]' '"<!-- coding-policy:claude-quality-review:v1 -->\nNo issues."')"
}
t_codex_marker() {
assert_eq "lane" "codex-policy" "$(lane_of 'github-actions[bot]' '"<!-- coding-policy:codex-policy-review:v1 -->\nPolicy loaded: 30 rule files."')"
}
t_unmarked_legacy_is_codex() {
assert_eq "plain body" "codex-policy" "$(lane_of 'github-actions[bot]' '"Policy loaded: 30 rule files."')" || return 1
assert_eq "null body" "codex-policy" "$(lane_of 'github-actions[bot]' 'null')"
}
# A marker anywhere but the first bytes never reclassifies a review.
t_quoted_marker_does_not_reclassify() {
assert_eq "lane" "codex-policy" "$(lane_of 'github-actions[bot]' '"Policy loaded. Finding quotes <!-- coding-policy:claude-quality-review:v1 -->"')"
}
# Copilot belongs to no lane, so its reviews are ignored.
t_copilot_login_has_no_lane() {
assert_eq "plain body" "null" "$(lane_of 'copilot-pull-request-reviewer[bot]' '"Copilot reviewed 3 files."')" || return 1
assert_eq "marked body" "null" "$(lane_of 'copilot-pull-request-reviewer[bot]' '"<!-- coding-policy:claude-quality-review:v1 -->"')"
}
# A marker from a login no lane trusts claims no lane.
t_untrusted_login_has_no_lane() {
assert_eq "human with claude marker" "null" "$(lane_of 'some-human' '"<!-- coding-policy:claude-quality-review:v1 -->"')" || return 1
assert_eq "human with codex marker" "null" "$(lane_of 'some-human' '"<!-- coding-policy:codex-policy-review:v1 -->"')"
}
echo "== review-lanes.sh tests =="
run "the executable emits the lane config" t_config_shape
run "the Claude workflow name matches review-claude.yml" t_claude_workflow_matches_the_shipped_workflow
run "a Claude marker under the shared login is Claude" t_claude_marker_under_shared_login
run "a Codex marker is Codex" t_codex_marker
run "an unmarked legacy review is Codex by login" t_unmarked_legacy_is_codex
run "a quoted marker does not reclassify" t_quoted_marker_does_not_reclassify
run "the Copilot login claims no lane" t_copilot_login_has_no_lane
run "an untrusted login claims no lane" t_untrusted_login_has_no_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