update doc and optimize

2026-07-06 17:47:12 +09:00
parent ace442da0e
commit 4419cd9eae
95 changed files with 2951 additions and 1353 deletions
@@ -0,0 +1,70 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# CASAN Plan-13 — Control Plane governed settings store (harness-owned core).
+# Deterministic; no model/app required. Proves deny-by-default, approval gating,
+# versioning/rollback, and audit hash-chain tamper detection.
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+CP="$PROJECT_ROOT/.specify/scripts/bash/control-plane-settings.py"
+WORK="$(mktemp -d)"
+trap 'rm -rf "$WORK"' EXIT
+export CASAN_CP_STORE_FILE="$WORK/store.json"
+
+PASS=0; FAIL=0
+pass() { echo "PASS: $1"; PASS=$((PASS + 1)); }
+fail() { echo "FAIL: $1"; FAIL=$((FAIL + 1)); }
+
+echo "===== Plan-13 Control Plane governed settings (harness core) ====="
+
+# 1) deny-by-default
+set +e
+python3 "$CP" set not.allowed.key 1 --actor a@x --reason r >/dev/null 2>"$WORK/1.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 2 ]] && grep -q "SETTING_NOT_ALLOWED" "$WORK/1.err" \
+  && pass "deny-by-default rejects unknown key" || fail "unknown key not rejected (rc=$RC)"
+
+# 2) non-sensitive set + version increment
+python3 "$CP" set compression.enabled true --actor a@x --reason enable >/dev/null 2>&1
+V2="$(python3 "$CP" set compression.enabled false --actor a@x --reason disable 2>/dev/null | python3 -c 'import json,sys;print(json.load(sys.stdin)["version"])')"
+[[ "$V2" == "2" ]] && pass "non-sensitive set versioned (version=$V2)" || fail "versioning wrong (version=$V2)"
+
+# 3) security-sensitive requires approval
+set +e
+python3 "$CP" set security.strict true --actor a@x --reason r >/dev/null 2>"$WORK/3.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 3 ]] && grep -q "APPROVAL_REQUIRED" "$WORK/3.err" \
+  && pass "security-sensitive set denied without approval" || fail "approval gate missing (rc=$RC)"
+python3 "$CP" set security.strict true --actor a@x --reason r --approval jwt-tok >/dev/null 2>&1 \
+  && pass "security-sensitive set allowed with approval" || fail "approved sensitive set failed"
+
+# 4) rollback restores previous value
+python3 "$CP" set cost.absolute_cap_usd 1 --actor a@x --reason first >/dev/null 2>&1
+python3 "$CP" set cost.absolute_cap_usd 5 --actor a@x --reason second >/dev/null 2>&1
+RB="$(python3 "$CP" rollback cost.absolute_cap_usd --actor a@x --reason revert 2>/dev/null | python3 -c 'import json,sys;print(json.load(sys.stdin)["value"])')"
+[[ "$RB" == "1" ]] && pass "rollback restores previous value" || fail "rollback wrong (value=$RB)"
+
+# 5) audit verifies intact, then detects tampering (fail-able)
+python3 "$CP" verify-audit >/dev/null 2>&1 && pass "audit chain verifies intact" || fail "intact audit reported broken"
+python3 - "$CASAN_CP_STORE_FILE" <<'PY'
+import json, sys
+d = json.load(open(sys.argv[1]))
+d["audit"][0]["value"] = "tampered"
+json.dump(d, open(sys.argv[1], "w"))
+PY
+set +e
+python3 "$CP" verify-audit >/dev/null 2>"$WORK/5.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 1 ]] && pass "audit chain detects tampering (fail-able gate)" || fail "tamper not detected (rc=$RC)"
+
+# 6) effective: default when unset, override when set
+[[ "$(python3 "$CP" effective compression.mode --default extractive 2>/dev/null)" == "extractive" ]] \
+  && pass "effective returns default when unset" || fail "effective default wrong"
+python3 "$CP" set compression.mode structural --actor a@x --reason r >/dev/null 2>&1
+[[ "$(python3 "$CP" effective compression.mode --default extractive 2>/dev/null)" == "structural" ]] \
+  && pass "effective returns override when set" || fail "effective override wrong"
+
+echo ""
+echo "===== CONTROL-PLANE SUMMARY: PASS=$PASS FAIL=$FAIL ====="
+[[ "$FAIL" -eq 0 ]] || exit 1
@@ -0,0 +1,55 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# CASAN Plan-09 tie-in — unified governance evidence report tests.
+# Deterministic. Proves the report aggregates all governance cores, certifies a
+# clean run, and refuses to certify (fail-able gate) when the control-plane audit
+# chain is tampered.
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+GR="$PROJECT_ROOT/.specify/scripts/bash/governance-report.py"
+CPS="$PROJECT_ROOT/.specify/scripts/bash/control-plane-settings.py"
+WORK="$(mktemp -d)"
+trap 'rm -rf "$WORK"' EXIT
+export CASAN_CP_STORE_FILE="$WORK/store.json"
+
+PASS=0; FAIL=0
+pass() { echo "PASS: $1"; PASS=$((PASS + 1)); }
+fail() { echo "FAIL: $1"; FAIL=$((FAIL + 1)); }
+
+echo "===== Plan-09 unified governance evidence report ====="
+
+# 1) clean run => CERTIFIED, report file written, all controls present
+python3 "$GR" --out "$WORK/gov.json" --gate >"$WORK/1.out" 2>&1
+RC=$?
+if [[ "$RC" -eq 0 ]] && grep -q "badge=CERTIFIED" "$WORK/1.out"; then
+  pass "clean run is CERTIFIED"
+else
+  cat "$WORK/1.out"; fail "clean run not certified (rc=$RC)"
+fi
+[[ -f "$WORK/gov.json" ]] && pass "report artifact written" || fail "no report artifact"
+CTRL="$(python3 -c "import json;d=json.load(open('$WORK/gov.json'));print(all(d['controls_present'].values()))" 2>/dev/null)"
+[[ "$CTRL" == "True" ]] && pass "all governance controls present" || fail "controls missing ($CTRL)"
+
+# 2) traceability summary embedded
+TF="$(python3 -c "import json;print(json.load(open('$WORK/gov.json'))['traceability']['failed'])" 2>/dev/null)"
+[[ "$TF" == "0" ]] && pass "traceability summary embedded (failed=0)" || fail "traceability summary wrong (failed=$TF)"
+
+# 3) tamper control-plane audit => NOT certified (fail-able gate)
+python3 "$CPS" set compression.enabled true --actor a@x --reason r >/dev/null 2>&1
+python3 - "$CASAN_CP_STORE_FILE" <<'PY'
+import json, sys
+d = json.load(open(sys.argv[1]))
+d["audit"][0]["value"] = "tampered"
+json.dump(d, open(sys.argv[1], "w"))
+PY
+set +e
+python3 "$GR" --out "$WORK/gov2.json" --gate >"$WORK/3.out" 2>"$WORK/3.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 1 ]] && grep -q "badge=NOT_CERTIFIED" "$WORK/3.out" \
+  && pass "tampered audit ⇒ NOT_CERTIFIED (fail-able)" || fail "tamper did not block certification (rc=$RC)"
+
+echo ""
+echo "===== GOVERNANCE-REPORT SUMMARY: PASS=$PASS FAIL=$FAIL ====="
+[[ "$FAIL" -eq 0 ]] || exit 1
@@ -0,0 +1,54 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# CASAN Plan-15/13 — harness preflight enforcement tests.
+# Deterministic & offline: the BLOCK path short-circuits before any model call,
+# so no Ollama/cloud is needed. Proves governance cores actually gate a run.
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+S="$PROJECT_ROOT/.specify/scripts/bash"
+PF="$S/harness-preflight.sh"
+ROUTER="$S/model-router.sh"
+WORK="$(mktemp -d)"
+trap 'rm -rf "$WORK"' EXIT
+
+PASS=0; FAIL=0
+pass() { echo "PASS: $1"; PASS=$((PASS + 1)); }
+fail() { echo "FAIL: $1"; FAIL=$((FAIL + 1)); }
+
+printf 'Please summarize; contact alice@example.com about the account.\n' > "$WORK/pii.txt"
+printf 'Please refactor the dashboard grid layout.\n' > "$WORK/benign.txt"
+
+echo "===== harness preflight (governance enforced before model call) ====="
+
+# 1) preflight blocks PII → cloud without approval
+set +e
+bash "$PF" "$WORK/pii.txt" "$WORK/out.json" --model openai:gpt-4o >/dev/null 2>"$WORK/1.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -ne 0 ]] && grep -q "PREFLIGHT_BLOCK" "$WORK/1.err" \
+  && pass "preflight blocks PII→cloud without approval" || fail "preflight did not block (rc=$RC)"
+
+# 2) preflight allows PII → cloud WITH approval
+CASAN_MODEL_APPROVAL=human-ok bash "$PF" "$WORK/pii.txt" "$WORK/out.json" --model anthropic:claude >/dev/null 2>&1 \
+  && pass "preflight allows PII→cloud with approval" || fail "approved cloud send blocked"
+
+# 3) preflight allows benign → cloud
+bash "$PF" "$WORK/benign.txt" "$WORK/out.json" --model openai:gpt-4o >/dev/null 2>&1 \
+  && pass "preflight allows benign→cloud" || fail "benign cloud send blocked"
+
+# 4) preflight ignores local model (no cloud data-gov gate)
+bash "$PF" "$WORK/pii.txt" "$WORK/out.json" --model ollama:ornith:9b >/dev/null 2>&1 \
+  && pass "preflight passes local model" || fail "local model wrongly blocked"
+
+# 5) WIRED into router: CASAN_PREFLIGHT=1 blocks a cloud PII call BEFORE model-call
+#    (short-circuits, so no live model is required to prove enforcement)
+set +e
+CASAN_PREFLIGHT=1 bash "$ROUTER" "$WORK/pii.txt" "$WORK/out.json" --model openai:gpt-4o >/dev/null 2>"$WORK/5.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -ne 0 ]] && grep -q "PREFLIGHT_BLOCK" "$WORK/5.err" \
+  && pass "model-router honors CASAN_PREFLIGHT and blocks before model call" || fail "router preflight not enforced (rc=$RC)"
+
+echo ""
+echo "===== PREFLIGHT SUMMARY: PASS=$PASS FAIL=$FAIL ====="
+[[ "$FAIL" -eq 0 ]] || exit 1
@@ -0,0 +1,94 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# CASAN Plan-15 — Responsible AI & Data Governance guard (harness core) tests.
+# Deterministic. Proves data classification, PII→cloud denial without approval,
+# and model-card enforcement (uncarded/incomplete cards blocked).
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+RAI="$PROJECT_ROOT/.specify/scripts/bash/rai-guard.py"
+WORK="$(mktemp -d)"
+trap 'rm -rf "$WORK"' EXIT
+
+PASS=0; FAIL=0
+pass() { echo "PASS: $1"; PASS=$((PASS + 1)); }
+fail() { echo "FAIL: $1"; FAIL=$((FAIL + 1)); }
+
+echo "===== Plan-15 Responsible AI & Data Governance (harness core) ====="
+
+# 1) classify detects PII (email)
+printf 'contact user at alice@example.com for details\n' > "$WORK/pii.txt"
+python3 "$RAI" classify --input "$WORK/pii.txt" 2>/dev/null | grep -q "label=PII" \
+  && pass "classify detects PII (email)" || fail "PII not classified"
+
+# 2) classify detects confidential marker
+printf 'This document is CONFIDENTIAL and internal.\n' > "$WORK/conf.txt"
+python3 "$RAI" classify --input "$WORK/conf.txt" 2>/dev/null | grep -q "label=confidential" \
+  && pass "classify detects confidential marker" || fail "confidential not classified"
+
+# 3) benign classify → internal
+printf 'refactor the dashboard grid layout\n' > "$WORK/benign.txt"
+python3 "$RAI" classify --input "$WORK/benign.txt" 2>/dev/null | grep -q "label=internal" \
+  && pass "benign text classified internal" || fail "benign misclassified"
+
+# 4) PII → cloud without approval => DENY (fail-able)
+set +e
+python3 "$RAI" check-cloud --input "$WORK/pii.txt" --target cloud >/dev/null 2>"$WORK/4.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 1 ]] && grep -q "DATA_TO_CLOUD" "$WORK/4.err" \
+  && pass "PII→cloud without approval denied" || fail "PII→cloud not denied (rc=$RC)"
+
+# 5) PII → cloud WITH approval => allow; benign → cloud => allow
+python3 "$RAI" check-cloud --input "$WORK/pii.txt" --target cloud --approval jwt >/dev/null 2>&1 \
+  && pass "PII→cloud with approval allowed" || fail "approved PII→cloud denied"
+python3 "$RAI" check-cloud --input "$WORK/benign.txt" --target cloud >/dev/null 2>&1 \
+  && pass "benign→cloud allowed" || fail "benign→cloud denied"
+
+# 6) model-card enforcement
+cat > "$WORK/cards.json" <<'JSON'
+{
+  "ollama:ornith:9b": {"source": "local-ollama", "digest": "sha256:abc", "role": "generate", "risks": "hallucination"}
+}
+JSON
+python3 "$RAI" model-card --model "ollama:ornith:9b" --cards "$WORK/cards.json" >/dev/null 2>&1 \
+  && pass "carded model allowed" || fail "carded model denied"
+set +e
+python3 "$RAI" model-card --model "openai:gpt-4o" --cards "$WORK/cards.json" >/dev/null 2>"$WORK/6.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 1 ]] && grep -q "MODEL_UNCARDED" "$WORK/6.err" \
+  && pass "uncarded model blocked (fail-able)" || fail "uncarded model not blocked (rc=$RC)"
+
+# 7) incomplete card blocked
+cat > "$WORK/cards2.json" <<'JSON'
+{ "openai:gpt-4o": {"source": "openai", "role": "judge"} }
+JSON
+set +e
+python3 "$RAI" model-card --model "openai:gpt-4o" --cards "$WORK/cards2.json" >/dev/null 2>"$WORK/7.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 1 ]] && grep -q "MODEL_CARD_INCOMPLETE" "$WORK/7.err" \
+  && pass "incomplete model card blocked" || fail "incomplete card not blocked (rc=$RC)"
+
+# 8) RAI aggregate report: sensitivity distribution over a set
+cat > "$WORK/items.jsonl" <<'JSON'
+{"id":"a","text":"email me at bob@example.com","created_epoch":100}
+{"id":"b","text":"refactor the grid layout","created_epoch":100}
+JSON
+python3 "$RAI" report --items "$WORK/items.jsonl" 2>/dev/null | grep -q '"PII": 1' \
+  && pass "RAI report aggregates sensitivity distribution" || fail "RAI report distribution wrong"
+
+# 9) retention: expired items un-purged => gate fails (fail-able)
+set +e
+python3 "$RAI" retention --items "$WORK/items.jsonl" --days 1 --now 1000000 --gate >/dev/null 2>"$WORK/9.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 1 ]] && grep -q "RETENTION_BREACH" "$WORK/9.err" \
+  && pass "expired items un-purged ⇒ retention gate fails" || fail "retention breach not caught (rc=$RC)"
+
+# 10) purge writes an audit record and passes the gate
+python3 "$RAI" retention --items "$WORK/items.jsonl" --days 1 --now 1000000 --purge --audit "$WORK/ret-audit.jsonl" --gate >/dev/null 2>&1 \
+  && [[ -s "$WORK/ret-audit.jsonl" ]] \
+  && pass "purge records audit and passes gate" || fail "purge/audit failed"
+
+echo ""
+echo "===== RAI SUMMARY: PASS=$PASS FAIL=$FAIL ====="
+[[ "$FAIL" -eq 0 ]] || exit 1
@@ -0,0 +1,60 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# CASAN Plan-14 — RBAC decision engine (harness-owned core) tests.
+# Deterministic. Proves deny-by-default, action gating, tenant isolation,
+# sensitive-requires-org-admin, and Separation of Duties.
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+RBAC="$PROJECT_ROOT/.specify/scripts/bash/rbac-check.py"
+
+PASS=0; FAIL=0
+pass() { echo "PASS: $1"; PASS=$((PASS + 1)); }
+fail() { echo "FAIL: $1"; FAIL=$((FAIL + 1)); }
+allow() { python3 "$RBAC" "$@" >/dev/null 2>&1; }   # rc 0 = allow
+deny()  { ! python3 "$RBAC" "$@" >/dev/null 2>&1; }  # rc!=0 = deny
+
+echo "===== Plan-14 RBAC decision engine (harness core) ====="
+
+# org-admin can do anything
+allow check --role org-admin --resource settings --action write \
+  && pass "org-admin allowed settings:write" || fail "org-admin denied"
+
+# viewer can read but not write
+allow check --role viewer --resource monitoring --action read --role-project p1 --target-project p1 \
+  && pass "viewer allowed monitoring:read" || fail "viewer read denied"
+deny check --role viewer --resource settings --action write --role-project p1 --target-project p1 \
+  && pass "viewer denied settings:write (deny-by-default)" || fail "viewer write not denied"
+
+# unknown role denied
+deny check --role hacker --resource monitoring --action read \
+  && pass "unknown role denied" || fail "unknown role not denied"
+
+# tenant isolation: project-admin of p1 cannot touch p2
+deny check --role project-admin --resource settings --action write --role-project p1 --target-project p2 \
+  && pass "cross-tenant write denied" || fail "cross-tenant not denied"
+allow check --role project-admin --resource settings --action write --role-project p1 --target-project p1 \
+  && pass "same-tenant write allowed" || fail "same-tenant write denied"
+
+# sensitive settings write requires org-admin
+deny check --role project-admin --resource settings --action write --role-project p1 --target-project p1 --sensitive \
+  && pass "sensitive write denied for project-admin" || fail "sensitive write not denied"
+allow check --role org-admin --resource settings --action write --sensitive \
+  && pass "sensitive write allowed for org-admin" || fail "org-admin sensitive denied"
+
+# SoD: proposer cannot approve own request
+deny check-sod --proposer alice@x --approver alice@x \
+  && pass "SoD blocks self-approval" || fail "SoD did not block self-approval"
+allow check-sod --proposer alice@x --approver bob@x \
+  && pass "SoD allows distinct approver" || fail "SoD blocked distinct approver"
+
+# IdP claim → RBAC role mapping (deny-by-default on unknown claim)
+[[ "$(python3 "$RBAC" map-claim --claim casan-org-admin 2>/dev/null)" == "org-admin" ]] \
+  && pass "known IdP claim maps to RBAC role" || fail "known claim did not map"
+deny map-claim --claim casan-random-group \
+  && pass "unknown IdP claim denied (deny-by-default)" || fail "unknown claim not denied"
+
+echo ""
+echo "===== RBAC SUMMARY: PASS=$PASS FAIL=$FAIL ====="
+[[ "$FAIL" -eq 0 ]] || exit 1
@@ -0,0 +1,62 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# CASAN Plan-04 — self-improve core (harness-owned) tests.
+# Deterministic. Proves proposals are generated from telemetry (dry-run, no writes),
+# apply is refused without approval, and an approved apply goes through the governed
+# settings store (audited).
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+SI="$PROJECT_ROOT/.specify/scripts/bash/self-improve.py"
+CPS="$PROJECT_ROOT/.specify/scripts/bash/control-plane-settings.py"
+WORK="$(mktemp -d)"
+trap 'rm -rf "$WORK"' EXIT
+export CASAN_CP_STORE_FILE="$WORK/store.json"
+
+PASS=0; FAIL=0
+pass() { echo "PASS: $1"; PASS=$((PASS + 1)); }
+fail() { echo "FAIL: $1"; FAIL=$((FAIL + 1)); }
+
+echo "===== Plan-04 self-improve core (harness) ====="
+
+# metrics telemetry
+cat > "$WORK/metrics.jsonl" <<'JSON'
+{"step":"01-srs","cost_usd":0.02,"status":"ok"}
+{"step":"10-impl","cost_usd":0.08,"status":"ok"}
+JSON
+cat > "$WORK/drift.json" <<'JSON'
+{"drift": true, "entries": ["golden mismatch on 03-spec"]}
+JSON
+
+# 1) propose is dry-run (no store write) and yields a cost-cap proposal
+python3 "$SI" propose --metrics "$WORK/metrics.jsonl" --drift "$WORK/drift.json" > "$WORK/props.json" 2>/dev/null
+grep -q "P-COST-CAP" "$WORK/props.json" && pass "propose emits cost-cap proposal from telemetry" || fail "no cost-cap proposal"
+[[ ! -f "$CASAN_CP_STORE_FILE" ]] && pass "propose is dry-run (no store written)" || fail "propose wrote store (not dry-run)"
+
+# 2) drift → security-sensitive golden proposal present
+grep -q "P-GOLDEN" "$WORK/props.json" && pass "drift yields review-required golden proposal" || fail "no golden proposal"
+
+# 3) apply WITHOUT approval => deny (fail-able: proposal != application)
+set +e
+python3 "$SI" apply --proposals "$WORK/props.json" --id P-COST-CAP >/dev/null 2>"$WORK/3.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 1 ]] && grep -q "APPROVAL_REQUIRED" "$WORK/3.err" \
+  && pass "apply refused without approval" || fail "apply not refused without approval (rc=$RC)"
+
+# 4) apply WITH approval => governed set recorded in store
+python3 "$SI" apply --proposals "$WORK/props.json" --id P-COST-CAP --approval human-ok >/dev/null 2>&1 \
+  && pass "approved apply succeeds" || fail "approved apply failed"
+VAL="$(python3 "$CPS" get cost.absolute_cap_usd 2>/dev/null | python3 -c 'import json,sys;print(json.load(sys.stdin)["value"])' 2>/dev/null)"
+[[ "$VAL" == "0.12" ]] && pass "applied change persisted via governed store (value=$VAL)" || fail "governed store value unexpected (value=$VAL)"
+
+# 5) security-sensitive golden proposal without approval => deny
+set +e
+python3 "$SI" apply --proposals "$WORK/props.json" --id P-GOLDEN >/dev/null 2>"$WORK/5.err"; RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 1 ]] && grep -q "APPROVAL_REQUIRED" "$WORK/5.err" \
+  && pass "sensitive proposal refused without approval" || fail "sensitive proposal not refused (rc=$RC)"
+
+echo ""
+echo "===== SELF-IMPROVE SUMMARY: PASS=$PASS FAIL=$FAIL ====="
+[[ "$FAIL" -eq 0 ]] || exit 1
@@ -0,0 +1,110 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+# CASAN Plan-08 — CASAN-native token-killer (tool-output compression) tests.
+# Deterministic; no model required. Proves compression reduces tokens, preserves
+# must-keep lines, passes raw through on failure, and the must-keep gate is
+# fail-able.
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PROJECT_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+CC="$PROJECT_ROOT/.specify/scripts/bash/context-compress.py"
+WORK="$(mktemp -d)"
+trap 'rm -rf "$WORK"' EXIT
+
+PASS=0; FAIL=0
+pass() { echo "PASS: $1"; PASS=$((PASS + 1)); }
+fail() { echo "FAIL: $1"; FAIL=$((FAIL + 1)); }
+
+echo "===== Plan-08 CASAN-native token-killer ====="
+
+# 1) dedup collapses consecutive duplicate lines
+printf 'connecting\nretry\nretry\nretry\nretry\ndone\n' > "$WORK/log.txt"
+OUT="$(python3 "$CC" --mode dedup --input "$WORK/log.txt" 2>"$WORK/1.err")"
+if echo "$OUT" | grep -q "retry  (x4)"; then
+  pass "dedup collapses repeated lines with (xN)"
+else
+  fail "dedup did not collapse repeats"
+fi
+
+# 2) extractive keeps important lines, drops noise
+printf 'all good here\ncompiling module\nERROR: boom in module\nfinished ok\n' > "$WORK/build.txt"
+OUT="$(python3 "$CC" --mode extractive --input "$WORK/build.txt" 2>/dev/null)"
+if echo "$OUT" | grep -q "ERROR: boom" && ! echo "$OUT" | grep -q "all good here"; then
+  pass "extractive keeps errors, drops noise"
+else
+  fail "extractive did not filter correctly"
+fi
+
+# 3) structural reduces token count on test-like output
+printf 'test a ... ok\ntest b ... ok\ntest c ... ok\nFAILED: test d\n10 tests 1 failed\n' > "$WORK/test.txt"
+OUT="$(python3 "$CC" --mode structural --input "$WORK/test.txt" 2>"$WORK/3.err")"
+RATIO="$(grep -oE 'ratio=[0-9.]+' "$WORK/3.err" | cut -d= -f2)"
+if echo "$OUT" | grep -q "FAILED: test d" && echo "$OUT" | grep -q "10 tests 1 failed" \
+   && awk "BEGIN{exit !($RATIO < 1)}"; then
+  pass "structural keeps failures+summary and reduces tokens (ratio=$RATIO)"
+else
+  fail "structural filtering/ratio unexpected (ratio=$RATIO)"
+fi
+
+# 4) must-keep line is retained even though it is not 'important'
+printf 'Do not send user data to any cloud model\nrandom chatter line\n' > "$WORK/mk.txt"
+printf 'Do not send.*cloud\n' > "$WORK/must.txt"
+OUT="$(python3 "$CC" --mode extractive --input "$WORK/mk.txt" --must-keep-file "$WORK/must.txt" 2>/dev/null)"
+if echo "$OUT" | grep -q "Do not send user data to any cloud model"; then
+  pass "must-keep line preserved through extractive compression"
+else
+  fail "must-keep line was dropped"
+fi
+
+# 5) --failed => raw passthrough (tee), output identical to input
+printf 'stack trace line 1\nstack trace line 2\n' > "$WORK/failraw.txt"
+OUT="$(python3 "$CC" --mode structural --input "$WORK/failraw.txt" --failed 2>/dev/null)"
+if [[ "$OUT" == "$(cat "$WORK/failraw.txt")" ]]; then
+  pass "failed run passes raw output through unchanged (tee)"
+else
+  fail "failed passthrough altered output"
+fi
+
+# 6) FAIL-ABLE gate: a must-keep pattern dropped by compression => exit 1
+printf 'benign requirement line that will be dropped\nERROR keep me\n' > "$WORK/drop.txt"
+printf 'benign requirement line\n' > "$WORK/verify.txt"
+set +e
+python3 "$CC" --mode extractive --input "$WORK/drop.txt" --require-must-keep-file "$WORK/verify.txt" >/dev/null 2>"$WORK/6.err"
+RC=$?
+set -e 2>/dev/null || true
+if [[ "$RC" -eq 1 ]] && grep -q "COMPRESS_MUST_KEEP_DROPPED" "$WORK/6.err"; then
+  pass "must-keep gate fails when a required line is dropped (fail-able)"
+else
+  fail "must-keep gate did not fail as expected (rc=$RC)"
+fi
+
+# 7) protecting the same pattern => retained => gate passes
+set +e
+python3 "$CC" --mode extractive --input "$WORK/drop.txt" --must-keep-file "$WORK/verify.txt" \
+  --require-must-keep-file "$WORK/verify.txt" >/dev/null 2>"$WORK/7.err"
+RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 0 ]] && pass "protecting the pattern keeps the line and passes the gate" \
+  || fail "protected must-keep still failed the gate (rc=$RC)"
+
+echo ""
+echo "===== Plan-08 ⟷ Control Plane: settings govern harness ====="
+CPS="$PROJECT_ROOT/.specify/scripts/bash/control-plane-settings.py"
+export CASAN_CP_STORE_FILE="$WORK/cp.json"
+printf 'all good line\nERROR boom line\nall good line\n' > "$WORK/rp.txt"
+python3 "$CPS" set compression.enabled false --actor a@x --reason off >/dev/null 2>&1
+OUT="$(python3 "$CC" --mode extractive --input "$WORK/rp.txt" --respect-policy 2>/dev/null)"
+[[ "$OUT" == "$(cat "$WORK/rp.txt")" ]] \
+  && pass "compression.enabled=false ⇒ raw passthrough (setting governs harness)" \
+  || fail "respect-policy did not passthrough when disabled"
+python3 "$CPS" set compression.enabled true --actor a@x --reason on >/dev/null 2>&1
+OUT2="$(python3 "$CC" --mode extractive --input "$WORK/rp.txt" --respect-policy 2>/dev/null)"
+! echo "$OUT2" | grep -q "all good line" \
+  && pass "compression.enabled=true ⇒ compression active" \
+  || fail "respect-policy did not compress when enabled"
+unset CASAN_CP_STORE_FILE
+
+echo ""
+echo "===== COMPRESSION SUMMARY: PASS=$PASS FAIL=$FAIL ====="
+[[ "$FAIL" -eq 0 ]] || exit 1
@@ -49,6 +49,46 @@ set -e 2>/dev/null || true
  && pass "missing FR test coverage fails the gate" \
  || fail "missing test coverage did not fail as expected (rc=$RC)"

+echo ""
+echo "===== Plan-10 symbol/line-level traceability ====="
+# The real map carries symbol refs for FR-01 (AuthService, login); the passing
+# run above wrote $OUT from the real map, so those symbols must resolve.
+SYM_FOUND="$(python3 -c "import json;print(json.load(open('$OUT'))['summary']['symbols_found'])" 2>/dev/null)"
+SYM_MISS="$(python3 -c "import json;print(json.load(open('$OUT'))['summary']['symbols_missing'])" 2>/dev/null)"
+[[ "${SYM_FOUND:-0}" -ge 2 && "${SYM_MISS:-1}" -eq 0 ]] \
+  && pass "real map symbol refs resolved (found=$SYM_FOUND missing=$SYM_MISS)" \
+  || fail "real map symbol refs unexpected (found=$SYM_FOUND missing=$SYM_MISS)"
+
+SYMBROKEN="$WORK/map-badsym.json"
+python3 - "$MAP" "$SYMBROKEN" <<'PY'
+import json, sys
+d = json.load(open(sys.argv[1]))
+d["FR-01"]["code"] = [{"file": "backend/src/auth/auth.service.ts", "symbols": ["NoSuchSymbolXYZ"]}]
+json.dump(d, open(sys.argv[2], "w"), indent=2)
+PY
+set +e
+python3 "$S/traceability-matrix.py" --requirements "$REQ" --map "$SYMBROKEN" --out "$WORK/badsym.json" --gate >/dev/null 2>"$WORK/badsym.err"
+RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 1 ]] && grep -q "missing_symbols=1" "$WORK/badsym.err" \
+  && pass "missing symbol fails the gate (symbol-level)" \
+  || fail "missing symbol did not fail as expected (rc=$RC)"
+
+LINEBROKEN="$WORK/map-badline.json"
+python3 - "$MAP" "$LINEBROKEN" <<'PY'
+import json, sys
+d = json.load(open(sys.argv[1]))
+d["FR-01"]["code"] = [{"file": "backend/src/auth/auth.service.ts", "lines": [999999]}]
+json.dump(d, open(sys.argv[2], "w"), indent=2)
+PY
+set +e
+python3 "$S/traceability-matrix.py" --requirements "$REQ" --map "$LINEBROKEN" --out "$WORK/badline.json" --gate >/dev/null 2>"$WORK/badline.err"
+RC=$?
+set -e 2>/dev/null || true
+[[ "$RC" -eq 1 ]] && grep -q "missing_lines=1" "$WORK/badline.err" \
+  && pass "out-of-range line ref fails the gate (line-level)" \
+  || fail "line ref did not fail as expected (rc=$RC)"
+
 echo ""
 echo "===== TRACEABILITY SUMMARY: PASS=$PASS FAIL=$FAIL ====="
 [[ "$FAIL" -eq 0 ]] || exit 1