thanhnvandClaude Opus 4.8
2c765c9a45
Task 1.2: introduce a single path resolver so no harness script hardcodes `.specify/...` scattered across the tree. casan-paths.sh resolves four roots (HARNESS/STATE/GOVERNANCE/APP) by marker-based walk-up from its own location — never `git rev-parse` (git root is the repo PARENT here, not the app dir). - 101 bash scripts/tests: 238 hardcoded `$PROJECT_ROOT/.specify/...` refs rewritten to CASAN_HARNESS_ROOT (code) / CASAN_STATE_ROOT (logs,state) / CASAN_GOVERNANCE_ROOT. Sandbox test vars ($WORK/$TP/$FP/$T1_WORK) left untouched. - Roots are NOT exported: each script/subprocess self-resolves from its own tree, matching the original per-script semantics and preserving hermetic sandbox isolation (node casan-step.mjs, copied telemetry/rollback scripts must not inherit real roots). - Sandbox tests that copy a harness script now also copy casan-paths.sh (its new sibling dependency): adversarial (verify-audit-chain/verify-tool-audit/rollback) + track-a (security-check/telemetry-integrity). - control-plane-settings.json reclassified as STATE (untracked runtime store). Roots all still resolve to `.specify` in this monolithic layout, so behavior is unchanged. Full gate: PASS=64 FAIL=0 SKIP=3 (adversarial 44/0, track-a 25/0). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
118 lines
4.5 KiB
Bash
118 lines
4.5 KiB
Bash
#!/usr/bin/env bash
|
|
set -uo pipefail
|
|
|
|
# CASAN Plan-17 Track 2 — No-progress / Oscillation Detector.
|
|
#
|
|
# Proves:
|
|
# * a strictly improving progress series -> CONVERGING (exit 0),
|
|
# * N identical consecutive actions -> OSCILLATING (repeat),
|
|
# * A,B,A,B... alternation -> OSCILLATING (thrash),
|
|
# * flat progress over the window -> STALLED,
|
|
# * on_stall=escalate -> ESCALATE (exit 4); on_stall=halt -> HALT (exit 3),
|
|
# * a corrupt observation stream fails closed (HALT, not silent CONVERGING),
|
|
# * verdicts write a hash-linked audit record,
|
|
# * state is redirected -> repo .specify/state stays clean.
|
|
#
|
|
# Deterministic; hermetic; no model / network / docker.
|
|
|
|
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
|
source "$SCRIPT_DIR/../scripts/bash/casan-paths.sh"
|
|
PROJECT_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
|
|
CONV="$CASAN_HARNESS_ROOT/scripts/bash/loop-convergence.py"
|
|
POLICY="$CASAN_HARNESS_ROOT/config/loop-policy.yaml"
|
|
|
|
WORK="$(mktemp -d)"
|
|
trap 'rm -rf "$WORK"' EXIT
|
|
export CASAN_LOOP_STATE_ROOT="$WORK/state"
|
|
export CASAN_LOOP_POLICY_FILE="$POLICY"
|
|
|
|
PASS=0; FAIL=0
|
|
pass() { echo "PASS: $1"; PASS=$((PASS + 1)); }
|
|
fail() { echo "FAIL: $1"; FAIL=$((FAIL + 1)); }
|
|
|
|
obs() { python3 "$CONV" observe --run-id "$1" --step "$2" --action-hash "$3" --progress "$4" >/dev/null; }
|
|
|
|
verdict() {
|
|
# verdict <expect_rc> <desc> <run-id> [profile]
|
|
local expect="$1" desc="$2" run="$3" prof="${4:-prod}"
|
|
set +e
|
|
OUT="$(python3 "$CONV" verdict --run-id "$run" --profile "$prof" 2>/dev/null)"
|
|
local rc=$?
|
|
set -e 2>/dev/null || true
|
|
if [[ "$rc" -eq "$expect" ]]; then pass "$desc (rc=$rc)"; else fail "$desc (rc=$rc, want $expect)"; fi
|
|
}
|
|
|
|
echo "===== Plan-17 T2: Convergence Detector ====="
|
|
|
|
# 1) improving progress -> CONVERGING
|
|
obs conv-ok 1 aaa 0.1; obs conv-ok 2 bbb 0.4; obs conv-ok 3 ccc 0.7; obs conv-ok 4 ddd 0.95
|
|
verdict 0 "improving progress -> CONVERGING" conv-ok
|
|
grep -q '"verdict": "CONVERGING"' <<<"$OUT" && pass "CONVERGING in envelope" || fail "missing CONVERGING"
|
|
|
|
# 2) N identical consecutive actions -> OSCILLATING (repeat). prod N=3.
|
|
obs osc-rep 1 same 0.2; obs osc-rep 2 same 0.3; obs osc-rep 3 same 0.4
|
|
verdict 4 "repeat action -> OSCILLATING (escalate)" osc-rep
|
|
grep -q '"verdict": "OSCILLATING"' <<<"$OUT" && pass "OSCILLATING (repeat) in envelope" || fail "missing OSCILLATING repeat"
|
|
|
|
# 3) A,B,A,B thrash over window (prod thrash_window=4) -> OSCILLATING
|
|
obs osc-thr 1 A 0.1; obs osc-thr 2 B 0.2; obs osc-thr 3 A 0.3; obs osc-thr 4 B 0.4
|
|
verdict 4 "thrash A,B,A,B -> OSCILLATING" osc-thr
|
|
grep -q '"pattern": "thrash"' <<<"$OUT" && pass "thrash pattern detected" || fail "thrash not detected"
|
|
|
|
# 4) flat progress over window (prod no_progress_window=3) -> STALLED
|
|
obs stall 1 a 0.5; obs stall 2 b 0.5; obs stall 3 c 0.5; obs stall 4 d 0.5
|
|
verdict 4 "flat progress -> STALLED (escalate)" stall
|
|
grep -q '"verdict": "STALLED"' <<<"$OUT" && pass "STALLED in envelope" || fail "missing STALLED"
|
|
|
|
# 5) on_stall=halt -> HALT (exit 3)
|
|
cat > "$WORK/halt-stall.yaml" <<'YAML'
|
|
version: 1
|
|
profiles:
|
|
prod:
|
|
convergence:
|
|
oscillation_repeat: 3
|
|
no_progress_window: 3
|
|
on_stall: halt
|
|
YAML
|
|
obs halt-run 1 x 0.5; obs halt-run 2 x 0.5; obs halt-run 3 x 0.5
|
|
CASAN_LOOP_POLICY_FILE="$WORK/halt-stall.yaml" verdict 3 "on_stall=halt -> HALT" halt-run
|
|
|
|
# 6) corrupt observation stream -> fail-closed HALT
|
|
mkdir -p "$CASAN_LOOP_STATE_ROOT/runs/broken"
|
|
printf 'not-json{{{\n' > "$CASAN_LOOP_STATE_ROOT/runs/broken/convergence.jsonl"
|
|
verdict 3 "corrupt observation stream -> fail-closed HALT" broken
|
|
grep -q '"reason": "fail_closed"' <<<"$OUT" && pass "corrupt stream fails closed" || fail "corrupt stream not fail-closed"
|
|
|
|
# 7) insufficient data -> CONVERGING (governor is the hard stop)
|
|
obs few 1 a 0.1
|
|
verdict 0 "insufficient data -> CONVERGING" few
|
|
|
|
# 8) verdict wrote a hash-linked audit record chaining to genesis
|
|
AUDIT="$CASAN_LOOP_STATE_ROOT/audit/loop-audit.jsonl"
|
|
if [[ -s "$AUDIT" ]]; then
|
|
pass "audit log written on verdict"
|
|
python3 - "$AUDIT" <<'PY' && pass "audit chain links to genesis" || fail "audit chain broken"
|
|
import json, sys
|
|
lines=[l for l in open(sys.argv[1]) if l.strip()]
|
|
prev="0"*64
|
|
for l in lines:
|
|
e=json.loads(l)
|
|
if e.get("prev_hash")!=prev: raise SystemExit(1)
|
|
prev=e.get("hash")
|
|
raise SystemExit(0 if lines else 1)
|
|
PY
|
|
else
|
|
fail "audit log not written"
|
|
fi
|
|
|
|
# 9) no repo pollution
|
|
if [[ -d "$CASAN_STATE_ROOT/state" ]] && [[ -n "$(ls -A "$CASAN_STATE_ROOT/state" 2>/dev/null)" ]]; then
|
|
fail "repo .specify/state was polluted"
|
|
else
|
|
pass "repo .specify/state stays clean"
|
|
fi
|
|
|
|
echo ""
|
|
echo "===== T2 SUMMARY: PASS=$PASS FAIL=$FAIL ====="
|
|
[[ "$FAIL" -eq 0 ]] || exit 1
|