claude-plugin-inference-arbitrage/tests/filing.test.sh

#!/usr/bin/env bash
# bin/filing-plan: output path A — the marker-based idempotency contract
# (FR-6.3), template rendering (FR-6.1), and repo-less degradation (FR-6.4).
#
# The load-bearing case is "run 1 files, run 2 comments". It is proved here the
# only way that means anything: run 1's ACTUAL rendered body is fed back as the
# tracker's state for run 2. Nothing is asserted about the marker by hand.
#
# tests/fixtures/gitea-issues.json is copied verbatim from real Gitea API
# responses (probe issue kotkan/claude-plugin-inference-arbitrage#9, since
# closed), so the shape this parses is the shape the API returns.
set -euo pipefail
cd "$(dirname "${BASH_SOURCE[0]}")/.."

BIN=bin/filing-plan
JUD=tests/fixtures/filing-judgments.json
tmp=$(mktemp -d)
trap 'rm -rf "$tmp"' EXIT
fail=0

check() {
	local what=$1 got=$2 want=$3
	if [ "$got" = "$want" ]; then
		echo "  ok   $what = $want"
	else
		echo "  FAIL $what = $got (want $want)" >&2
		fail=1
	fi
}
# q <file> <python-expr over `d`>
q() { python3 -c "import json,sys;d=json.load(open(sys.argv[1]));print($2)" "$1"; }

bin/boundary-classify "$JUD" >"$tmp/cls.json"

plan() { # plan <existing.json> <out.json> [extra args...]
	local existing=$1 out=$2
	shift 2
	$BIN plan --classified "$tmp/cls.json" --judgments "$JUD" \
		--target-repo oleks/claude-plugin-fixture --existing "$existing" \
		--audit-page Audits/fixture-plugin/2026-07-29 \
		--window "2026-07-22 -> 2026-07-29" --run-date 2026-07-29 "$@" >"$out"
}

echo "== the fixture's two candidates grade as expected (guards the rest) =="
check "A verdict" "$(q "$tmp/cls.json" "d['candidates'][0]['verdict']")" "file"
check "B verdict" "$(q "$tmp/cls.json" "d['candidates'][1]['verdict']")" "no-offload"

echo "== RUN 1: empty tracker -> files, with the marker as line 1 =="
echo '[]' >"$tmp/empty.json"
plan "$tmp/empty.json" "$tmp/plan1.json"
check "to_file" "$(q "$tmp/plan1.json" "d['summary']['to_file']")" "1"
check "to_comment" "$(q "$tmp/plan1.json" "d['summary']['to_comment']")" "0"
check "action" "$(q "$tmp/plan1.json" "d['actions'][0]['action']")" "file"
check "marker is line 1" \
	"$(q "$tmp/plan1.json" "d['actions'][0]['body'].splitlines()[0]")" \
	"<!-- ia-candidate: fixture-plugin/skill-a/label-derivation -->"
check "token-offload label required" \
	"$(q "$tmp/plan1.json" "d['actions'][0]['required_labels']")" "['token-offload']"
check "non-file verdict skipped" "$(q "$tmp/plan1.json" "d['actions'][1]['action']")" "skip"

echo "== the body carries every element of the FR-6 contract =="
body=$(q "$tmp/plan1.json" "d['actions'][0]['body']")
for needle in \
	"**Measured cost.**" "**Evidence strength.**" "**Boundary position.**" \
	"Determinism tests: T1 ✓ T2 ✓ T3 ✓ T4 ✓ T5 ✓" \
	"**Proposed signature.**" "derive_labels(" "**Examples.**" \
	"**When a human would overrule this.**" "**Escalation path.**" \
	"**What crosses the boundary.**" \
	"Audits/fixture-plugin/2026-07-29" "Rubric v1.0.0"; do
	if grep -qF -- "$needle" <<<"$body"; then
		echo "  ok   body contains: $needle"
	else
		echo "  FAIL body missing: $needle" >&2
		fail=1
	fi
done
check "measurement is quoted, not invented" \
	"$(grep -cF 'Recurring sequence observed 17x' <<<"$body")" "1"
check "3 examples incl. the edge case" "$(grep -c '^- ' <<<"$body")" "3"

echo "== RUN 2 (FR-6.3, S5): run 1's own body is now on the tracker -> comment =="
python3 - "$tmp/plan1.json" "$tmp/tracker.json" <<'PY'
import json, sys
plan = json.load(open(sys.argv[1]))
filed = next(a for a in plan["actions"] if a["action"] == "file")
json.dump([{"number": 41, "state": "open", "title": filed["proposed_title"],
            "body": filed["body"], "comments": []}], open(sys.argv[2], "w"))
PY
plan "$tmp/tracker.json" "$tmp/plan2.json" --status grown
check "files nothing on re-run" "$(q "$tmp/plan2.json" "d['summary']['to_file']")" "0"
check "comments instead" "$(q "$tmp/plan2.json" "d['summary']['to_comment']")" "1"
check "on the right issue" \
	"$(q "$tmp/plan2.json" "d['actions'][0]['issue']")" "oleks/claude-plugin-fixture#41"
check "comment re-carries the marker" \
	"$(q "$tmp/plan2.json" "d['actions'][0]['body'].splitlines()[0]")" \
	"<!-- ia-candidate: fixture-plugin/skill-a/label-derivation -->"
check "comment reports the diff status" \
	"$(q "$tmp/plan2.json" "'Status: grown' in d['actions'][0]['body']")" "True"

echo "== RUN 3: the comment from run 2 is the only marker left (body rewritten) =="
python3 - "$tmp/plan2.json" "$tmp/tracker3.json" <<'PY'
import json, sys
plan = json.load(open(sys.argv[1]))
c = next(a for a in plan["actions"] if a["action"] == "comment")
json.dump([{"number": 41, "state": "closed", "title": "a human retitled this",
            "body": "and rewrote the body, dropping the marker",
            "comments": [{"body": c["body"]}]}], open(sys.argv[2], "w"))
PY
plan "$tmp/tracker3.json" "$tmp/plan3.json"
check "still no duplicate" "$(q "$tmp/plan3.json" "d['summary']['to_file']")" "0"
check "found via the comment" "$(q "$tmp/plan3.json" "d['actions'][0]['action']")" "comment"
check "closed state surfaced for issuer" \
	"$(q "$tmp/plan3.json" "d['actions'][0]['issue_state']")" "closed"

echo "== real Gitea API shapes parse, and a DIFFERENT id does not false-match =="
plan tests/fixtures/gitea-issues.json "$tmp/plan4.json"
check "unrelated markers -> files" "$(q "$tmp/plan4.json" "d['summary']['to_file']")" "1"
check "searched the fixture" "$(q "$tmp/plan4.json" "d['summary']['searched_issues']")" "1"

echo "== a marker on two issues warns and takes the lowest number =="
python3 - "$tmp/plan1.json" "$tmp/dupes.json" <<'PY'
import json, sys
filed = next(a for a in json.load(open(sys.argv[1]))["actions"] if a["action"] == "file")
json.dump([{"number": 77, "state": "open", "title": "later dupe", "body": filed["body"], "comments": []},
           {"number": 41, "state": "open", "title": "original", "body": filed["body"], "comments": []}],
          open(sys.argv[2], "w"))
PY
plan "$tmp/dupes.json" "$tmp/plan5.json"
check "lowest issue wins" "$(q "$tmp/plan5.json" "d['actions'][0]['issue_number']")" "41"
check "warns a human" "$(q "$tmp/plan5.json" "len(d['warnings'])")" "1"

echo "== --existing is mandatory: a skipped search must never mean 'file it all' =="
if $BIN plan --classified "$tmp/cls.json" --judgments "$JUD" \
	--target-repo oleks/claude-plugin-fixture --audit-page A --window w \
	--run-date 2026-07-29 >/dev/null 2>&1; then
	echo "  FAIL plan ran without --existing" >&2
	fail=1
else
	echo "  ok   refuses to plan without a search result"
fi

echo "== FR-6.4: a target with no writable repo degrades to summary-only =="
# Real, not synthetic: third-party/baoyu-design is a vendored plugin whose only
# remote is github.com, i.e. nothing this environment can file to.
THIRD=../third-party/baoyu-design
if [ -d "$THIRD" ]; then
	bin/plugin-inventory "$THIRD" --json >"$tmp/tp-inv.json"
	check "inventory still resolves it" "$(q "$tmp/tp-inv.json" "d['plugin']['name']")" "baoyu-design"
	$BIN queries --classified "$tmp/cls.json" --judgments "$JUD" \
		--target-path "$THIRD" --workspace /nonexistent >"$tmp/tp-q.json"
	check "filing" "$(q "$tmp/tp-q.json" "d['filing']")" "unavailable"
	check "reason names the remote" \
		"$(q "$tmp/tp-q.json" "'github.com' in d['filing_unavailable_reason']")" "True"
	check "no queries emitted" "$(q "$tmp/tp-q.json" "'queries' in d")" "False"

	$BIN plan --classified "$tmp/cls.json" --judgments "$JUD" \
		--target-path "$THIRD" --workspace /nonexistent --existing "$tmp/empty.json" \
		--audit-page A --window w --run-date 2026-07-29 >"$tmp/tp-plan.json"
	check "nothing filed" "$(q "$tmp/tp-plan.json" "d['summary']['to_file']")" "0"
	check "candidate marked unavailable" \
		"$(q "$tmp/tp-plan.json" "d['actions'][0]['filing']")" "unavailable"
else
	echo "  skip (no third-party/baoyu-design checkout)"
fi

echo "== a writable target resolves through its own checkout's remote =="
if [ -d ../worktree-discipline/.git ]; then
	$BIN queries --classified "$tmp/cls.json" --judgments "$JUD" \
		--target-path ../worktree-discipline >"$tmp/wd-q.json"
	check "repo" "$(q "$tmp/wd-q.json" "d['repo']")" "kotkan/claude-plugin-worktree-discipline"
	check "filing" "$(q "$tmp/wd-q.json" "d['filing']")" "available"
	check "label scan is the authoritative query" \
		"$(q "$tmp/wd-q.json" "d['queries']['label_scan']['labels']")" "['token-offload']"
else
	echo "  skip (no sibling worktree-discipline checkout)"
fi

exit "$fail"