qualia-framework 4.1.1 → 4.4.0

package/tests/lib.test.sh

@@ -0,0 +1,217 @@
+#!/bin/bash
+# Qualia Framework — bin/plan-contract.js + bin/agent-runs.js
+# Run: bash tests/lib.test.sh
+
+PASS=0
+FAIL=0
+FRAMEWORK_DIR="$(cd "$(dirname "$0")/.." && pwd)"
+NODE="${NODE:-node}"
+
+TMP_DIRS=()
+cleanup() {
+  for d in "${TMP_DIRS[@]}"; do
+    [ -d "$d" ] && rm -rf "$d"
+  done
+}
+trap cleanup EXIT
+
+mktmp() {
+  local t
+  t=$(mktemp -d)
+  TMP_DIRS+=("$t")
+  echo "$t"
+}
+
+ok()   { echo "  ✓ $1"; PASS=$((PASS + 1)); }
+fail() { echo "  ✗ $1"; FAIL=$((FAIL + 1)); }
+
+echo "lib.test.sh — plan-contract + agent-runs"
+
+# ─── plan-contract ──────────────────────────────────────────
+
+PC="$FRAMEWORK_DIR/bin/plan-contract.js"
+$NODE --check "$PC" >/dev/null 2>&1 && ok "plan-contract.js parses" || fail "plan-contract.js parse"
+
+OUT=$($NODE -e '
+const pc = require("'"$PC"'");
+const good = {
+  version: 1, phase: 1, goal: "x", why: "y",
+  generated_at: "2026-04-28T12:00:00Z", generated_by: "planner",
+  source_plan_hash: "sha256:abc", success_criteria: ["sc"],
+  tasks: [{
+    id: "T1", title: "t1", wave: 1, depends_on: [],
+    files_modify: [], files_create: [], files_delete: [],
+    acceptance_criteria: ["ac"], action: "do", context_files: [],
+    verification: [{ type: "file-exists", path: "a.ts" }]
+  }]
+};
+const errs = pc.validate(good);
+console.log(errs.length === 0 ? "VALID" : "INVALID:" + errs.join(";"));
+')
+[ "$OUT" = "VALID" ] && ok "validates well-formed contract" || fail "well-formed contract: $OUT"
+
+OUT=$($NODE -e '
+const pc = require("'"$PC"'");
+const errs = pc.validate({ version: 2, phase: 0, tasks: [] });
+console.log(errs.length > 0 ? "REJECTED" : "ACCEPTED");
+')
+[ "$OUT" = "REJECTED" ] && ok "rejects malformed contract" || fail "malformed accepted"
+
+OUT=$($NODE -e '
+const pc = require("'"$PC"'");
+const c = {
+  version: 1, phase: 1, goal: "x", why: "y",
+  generated_at: "t", generated_by: "planner", source_plan_hash: "h",
+  success_criteria: ["sc"],
+  tasks: [
+    { id: "T1", title: "a", wave: 1, depends_on: ["T2"], files_modify: [], files_create: [], files_delete: [],
+      acceptance_criteria: ["x"], action: "", context_files: [],
+      verification: [{ type: "file-exists", path: "a" }] },
+    { id: "T2", title: "b", wave: 1, depends_on: ["T1"], files_modify: [], files_create: [], files_delete: [],
+      acceptance_criteria: ["x"], action: "", context_files: [],
+      verification: [{ type: "file-exists", path: "b" }] },
+  ],
+};
+const errs = pc.validate(c);
+const cycle = errs.some(e => /cycle/.test(e));
+console.log(cycle ? "CYCLE-DETECTED" : "MISSED:" + errs.join(";"));
+')
+[ "$OUT" = "CYCLE-DETECTED" ] && ok "detects depends_on cycle" || fail "cycle missed: $OUT"
+
+OUT=$($NODE -e '
+const pc = require("'"$PC"'");
+const c = {
+  version: 1, phase: 1, goal: "x", why: "y",
+  generated_at: "t", generated_by: "planner", source_plan_hash: "h",
+  success_criteria: ["sc"],
+  tasks: [{ id: "T1", title: "a", wave: 1, depends_on: [], files_modify: [], files_create: [], files_delete: [],
+    acceptance_criteria: ["x"], action: "", context_files: [],
+    verification: [{ type: "command-exit", command: "node && rm", args: [], expected_exit: 0 }] }],
+};
+const errs = pc.validate(c);
+const blocked = errs.some(e => /shell metacharacters/.test(e));
+console.log(blocked ? "BLOCKED" : "ALLOWED");
+')
+[ "$OUT" = "BLOCKED" ] && ok "blocks shell metacharacters in command-exit" || fail "shell metas allowed: $OUT"
+
+OUT=$($NODE -e '
+const pc = require("'"$PC"'");
+const c = {
+  version: 1, phase: 1, goal: "x", why: "y",
+  generated_at: "t", generated_by: "planner", source_plan_hash: "h",
+  success_criteria: ["sc"],
+  tasks: [{ id: "T1", title: "a", wave: 1, depends_on: [], files_modify: [], files_create: [], files_delete: [],
+    acceptance_criteria: ["x"], action: "", context_files: [],
+    verification: [{ type: "behavioral", description: "feel", evidence_required: [] }] }],
+};
+const errs = pc.validate(c);
+const blocked = errs.some(e => /evidence_required/.test(e));
+console.log(blocked ? "BLOCKED" : "ALLOWED");
+')
+[ "$OUT" = "BLOCKED" ] && ok "rejects behavioral check with empty evidence" || fail "empty evidence allowed: $OUT"
+
+# Drift detection
+TMP=$(mktmp)
+$NODE -e '
+const pc = require("'"$PC"'");
+const fs = require("fs"), path = require("path");
+const dir = "'"$TMP"'/.planning";
+fs.mkdirSync(dir, { recursive: true });
+fs.writeFileSync(path.join(dir, "phase-1-plan.md"), "plan v1");
+const h = pc.hashPlan("plan v1");
+fs.writeFileSync(path.join(dir, "phase-1-contract.json"), JSON.stringify({ source_plan_hash: h }));
+const d1 = pc.checkDrift(path.join(dir, "phase-1-contract.json"), path.join(dir, "phase-1-plan.md"));
+console.log(d1.drift ? "FAIL1" : "");
+fs.appendFileSync(path.join(dir, "phase-1-plan.md"), "\nedit\n");
+const d2 = pc.checkDrift(path.join(dir, "phase-1-contract.json"), path.join(dir, "phase-1-plan.md"));
+console.log(d2.drift ? "DRIFT-DETECTED" : "FAIL2");
+' | tail -1 > /tmp/qfdrift.out
+[ "$(cat /tmp/qfdrift.out)" = "DRIFT-DETECTED" ] && ok "drift detection" || fail "drift detection: $(cat /tmp/qfdrift.out)"
+
+# ─── agent-runs ──────────────────────────────────────────
+
+AR="$FRAMEWORK_DIR/bin/agent-runs.js"
+$NODE --check "$AR" >/dev/null 2>&1 && ok "agent-runs.js parses" || fail "agent-runs.js parse"
+
+TMP=$(mktmp)
+mkdir -p "$TMP/.planning"
+RES=$($NODE -e '
+const ar = require("'"$AR"'");
+const tok = ar.start({ agent_type: "builder", model: "claude-sonnet-4-6", phase: 1, task_id: "T1" });
+const r = ar.finish(tok, { status: "success", commit_sha: "abc", cwd: "'"$TMP"'" });
+console.log(r.written ? "WRITTEN" : "NOT");
+')
+[ "$RES" = "WRITTEN" ] && ok "agent-runs writes a record" || fail "agent-runs write: $RES"
+
+RES=$($NODE -e '
+const ar = require("'"$AR"'");
+const recs = ar.read("'"$TMP"'");
+console.log(recs.length === 1 && recs[0].agent_type === "builder" ? "READ-OK" : "READ-FAIL");
+')
+[ "$RES" = "READ-OK" ] && ok "agent-runs reads back the record" || fail "read failed: $RES"
+
+# Telemetry off → no write
+TMP2=$(mktmp)
+mkdir -p "$TMP2/.planning"
+RES=$(QUALIA_TELEMETRY=off $NODE -e '
+const ar = require("'"$AR"'");
+const tok = ar.start({ agent_type: "verifier", model: "claude-opus-4-7" });
+const r = ar.finish(tok, { status: "success", cwd: "'"$TMP2"'" });
+console.log(r.written ? "WRITTEN" : "SKIPPED");
+')
+[ "$RES" = "SKIPPED" ] && ok "QUALIA_TELEMETRY=off disables writes" || fail "telemetry off: $RES"
+
+# Read still works after opt-out (history not hidden) — synthesize one record then disable
+TMP3=$(mktmp)
+mkdir -p "$TMP3/.planning"
+$NODE -e '
+const ar = require("'"$AR"'");
+const tok = ar.start({ agent_type: "builder", model: "x" });
+ar.finish(tok, { status: "success", cwd: "'"$TMP3"'" });
+' >/dev/null
+RES=$(QUALIA_TELEMETRY=off $NODE -e '
+const ar = require("'"$AR"'");
+console.log(ar.read("'"$TMP3"'").length);
+')
+[ "$RES" = "1" ] && ok "QUALIA_TELEMETRY=off does NOT hide existing records" || fail "history hidden: $RES"
+
+# Failure log file
+TMP4=$(mktmp)
+mkdir -p "$TMP4/.planning"
+RES=$($NODE -e '
+const ar = require("'"$AR"'");
+const tok = ar.start({ agent_type: "verifier", model: "x" });
+const r = ar.finish(tok, { status: "failure", failure_reason: "tsc-failed",
+  failure_detail: "x".repeat(800), full_stderr: "trace lines", cwd: "'"$TMP4"'" });
+console.log(r.log_file ? "LOG:" + r.log_file : "NO-LOG");
+' )
+echo "$RES" | grep -q '^LOG:' && ok "failure writes side log file" || fail "no log on failure: $RES"
+
+# Truncation: must keep tail
+RES=$($NODE -e '
+const ar = require("'"$AR"'");
+const recs = ar.read("'"$TMP4"'");
+const fd = recs[0].failure_detail;
+const allXs = /^x+$/.test(fd) && fd.length === 500;
+console.log(allXs ? "TAIL-KEPT-500" : "WRONG-LEN-OR-CONTENT:" + fd.length);
+')
+[ "$RES" = "TAIL-KEPT-500" ] && ok "failure_detail keeps tail at 500 chars" || fail "truncation: $RES"
+
+# Prune
+TMP5=$(mktmp)
+mkdir -p "$TMP5/.planning"
+$NODE -e '
+const ar = require("'"$AR"'");
+const t1 = ar.start({ agent_type: "builder", model: "x" });
+ar.finish(t1, { status: "success", cwd: "'"$TMP5"'" });
+' >/dev/null
+RES=$($NODE -e '
+const ar = require("'"$AR"'");
+const r = ar.prune("'"$TMP5"'", "2099-01-01T00:00:00Z");
+console.log("removed:" + r.removed);
+')
+[ "$RES" = "removed:1" ] && ok "prune --before removes old records" || fail "prune: $RES"
+
+echo ""
+echo "lib.test.sh: $PASS passed, $FAIL failed"
+[ "$FAIL" -eq 0 ]