fix(e2e): redact replay evidence and clean up compose resources

Makefile

@@ -48,6 +48,7 @@ harness-check: ## Validate the Bub replay harness and committed scenarios.
 	@uv run ruff check e2e/bub
 	@uv run ruff format --check e2e/bub
 	@uv run ty check --project e2e/bub --python e2e/bub/.venv e2e/bub/src integrations/bub/src
+	@uv run --project e2e/bub python -m pytest e2e/bub/tests
 	@uv run --project e2e/bub powercontext-e2e --help >/dev/null
 
 .PHONY: harness-acceptance

e2e/bub/README.md

@@ -51,4 +51,5 @@ POWERCONTEXT_E2E_DATABASE=oceanbase make harness-compose-acceptance
 ```
 
 `make harness-compose-live` uses the provider variables above. Evidence is written below `.powercontext-e2e/bub/`;
-set `POWERCONTEXT_E2E_OUTPUT` to keep it elsewhere. `make harness-compose-down` removes containers and database volumes.
+set `POWERCONTEXT_E2E_OUTPUT` to keep it elsewhere. Compose containers, networks, and volumes are removed after both
+successful and failed runs. `make harness-compose-down` remains available as an idempotent manual cleanup.

e2e/bub/pyproject.toml

@@ -13,6 +13,9 @@ dependencies = [
     "pyyaml>=6,<7",
 ]
 
+[dependency-groups]
+dev = ["pytest>=9.0.2"]
+
 [project.scripts]
 powercontext-e2e = "powercontext_e2e.__main__:main"
 
@@ -38,5 +41,8 @@ line-length = 120
 select = ["A", "B", "C4", "C90", "E", "F", "I", "PGH", "RUF", "S", "SIM", "T10", "TRY", "UP", "W", "YTT"]
 ignore = ["E501"]
 
+[tool.ruff.lint.per-file-ignores]
+"tests/**/*.py" = ["S101"]
+
 [tool.ruff.format]
 preview = true

e2e/bub/run.sh

@@ -68,6 +68,26 @@ if [ -z "${GITHUB_SHA:-}" ]; then
     export GITHUB_SHA
 fi
 
+cleanup() {
+    status=$?
+    trap - EXIT INT TERM
+    set +e
+
+    docker compose $compose_files down --volumes --remove-orphans
+    cleanup_status=$?
+    if [ "$cleanup_status" -ne 0 ]; then
+        echo "Compose cleanup failed with exit code $cleanup_status" >&2
+        if [ "$status" -eq 0 ]; then
+            status=$cleanup_status
+        fi
+    fi
+    exit "$status"
+}
+
+trap cleanup EXIT
+trap 'exit 130' INT
+trap 'exit 143' TERM
+
 docker compose $compose_files build powercontext harness
 docker compose $compose_files up --detach --wait powercontext
 
@@ -81,5 +101,3 @@ else
     scenario=${POWERCONTEXT_E2E_SCENARIO:-e2e/bub/scenarios/project-database-decision.yaml}
     docker compose $compose_files run --rm harness live "$scenario" --output /evidence
 fi
-
-docker compose $compose_files down --volumes --remove-orphans

e2e/bub/src/powercontext_e2e/redaction.py

@@ -0,0 +1,190 @@
+"""Credential redaction at the replay evidence boundary."""
+
+from __future__ import annotations
+
+import os
+import re
+from collections.abc import Mapping
+from dataclasses import dataclass
+from typing import Any
+from urllib.parse import parse_qsl, quote, unquote, urlsplit
+
+REDACTED = "[REDACTED]"
+
+_SENSITIVE_ENVIRONMENT_NAMES = frozenset({
+    "api_key",
+    "authorization",
+    "credentials",
+    "database_url",
+    "mysql_pwd",
+    "password",
+    "pgpassword",
+    "private_key",
+    "secret",
+    "token",
+})
+_SENSITIVE_ENVIRONMENT_SUFFIXES = (
+    "_access_key_id",
+    "_api_key",
+    "_authorization",
+    "_client_secret",
+    "_credentials",
+    "_database_url",
+    "_dsn",
+    "_password",
+    "_private_key",
+    "_secret",
+    "_secret_access_key",
+    "_secret_key",
+    "_token",
+)
+_SENSITIVE_FIELD_NAMES = frozenset({
+    "access_token",
+    "api_key",
+    "auth_token",
+    "authorization",
+    "client_secret",
+    "connection_string",
+    "credentials",
+    "database_url",
+    "dsn",
+    "password",
+    "private_key",
+    "proxy_authorization",
+    "refresh_token",
+    "secret",
+    "secret_key",
+    "token",
+})
+_SENSITIVE_FIELD_SUFFIXES = (
+    "_access_token",
+    "_api_key",
+    "_auth_token",
+    "_authorization",
+    "_client_secret",
+    "_connection_string",
+    "_credentials",
+    "_database_url",
+    "_dsn",
+    "_password",
+    "_private_key",
+    "_refresh_token",
+    "_secret",
+    "_secret_key",
+)
+_DATABASE_URL = re.compile(
+    r"(?i)\b(?:cockroachdb|mariadb|mssql|mysql|oceanbase|oracle|postgres|postgresql|sqlite)"
+    r"(?:\+[a-z0-9_.-]+)?://[^\s\"'`<>{}\[\](),;\\]+"
+)
+_URL_PASSWORD = re.compile(r"(?i)(\b[a-z][a-z0-9+.-]*://[^/\s:@]+:)[^@/\s\"'`<>]+(@)")
+_URL_QUERY_CREDENTIAL = re.compile(
+    r"(?i)([?&](?:access[_-]?token|api[_-]?key|auth[_-]?token|client[_-]?secret|password|refresh[_-]?token|secret|token)=)"
+    r"[^&#\s\"'`]+"
+)
+_CREDENTIAL_ASSIGNMENT = re.compile(
+    r"(?i)(\b(?:access[_-]?token|api[_-]?key|auth[_-]?token|authorization|client[_-]?secret|password|secret)"
+    r"\b\s*[:=]\s*)(?:(?:basic|bearer)\s+)?[^\s,;&\"'`]+"
+)
+_CREDENTIAL_OPTION = re.compile(
+    r"(?i)(--(?:access-token|api-key|auth-token|client-secret|password|token)\s+)"
+    r"[^\s,;\"'`]+"
+)
+_BEARER_CREDENTIAL = re.compile(r"(?i)\bbearer\s+[a-z0-9._~+/=-]{16,}")
+_PROVIDER_CREDENTIAL = re.compile(r"(?<![a-zA-Z0-9_-])sk-[a-zA-Z0-9_-]{8,}(?![a-zA-Z0-9_-])")
+
+
+@dataclass(frozen=True)
+class EvidenceRedactor:
+    """Remove configured and structurally recognizable credentials from evidence."""
+
+    secrets: tuple[str, ...]
+
+    @classmethod
+    def from_environment(cls) -> EvidenceRedactor:
+        secrets: set[str] = set()
+        for name, value in os.environ.items():
+            if value and _is_sensitive_environment_name(name):
+                secrets.update(_secret_variants(value))
+        return cls(tuple(sorted(secrets, key=lambda item: (-len(item), item))))
+
+    def redact(self, value: Any) -> Any:
+        """Recursively redact JSON-compatible evidence without changing its normal shape."""
+        return self._redact(value, sensitive=False)
+
+    def redact_text(self, value: str) -> str:
+        """Redact one final serialized evidence value."""
+        for secret in self.secrets:
+            if len(secret) >= 8:
+                value = value.replace(secret, REDACTED)
+            else:
+                value = re.sub(rf"(?<![a-zA-Z0-9]){re.escape(secret)}(?![a-zA-Z0-9])", REDACTED, value)
+        value = _DATABASE_URL.sub(REDACTED, value)
+        value = _URL_PASSWORD.sub(rf"\1{REDACTED}\2", value)
+        value = _URL_QUERY_CREDENTIAL.sub(rf"\1{REDACTED}", value)
+        value = _CREDENTIAL_ASSIGNMENT.sub(rf"\1{REDACTED}", value)
+        value = _CREDENTIAL_OPTION.sub(rf"\1{REDACTED}", value)
+        value = _PROVIDER_CREDENTIAL.sub(REDACTED, value)
+        return _BEARER_CREDENTIAL.sub(REDACTED, value)
+
+    def _redact(self, value: Any, *, sensitive: bool) -> Any:
+        if isinstance(value, str):
+            return REDACTED if sensitive and value else self.redact_text(value)
+        if isinstance(value, Mapping):
+            return {
+                self.redact_text(key) if isinstance(key, str) else key: self._redact(
+                    item,
+                    sensitive=sensitive or (isinstance(key, str) and _is_sensitive_field_name(key)),
+                )
+                for key, item in value.items()
+            }
+        if isinstance(value, list):
+            return [self._redact(item, sensitive=sensitive) for item in value]
+        if isinstance(value, tuple):
+            return tuple(self._redact(item, sensitive=sensitive) for item in value)
+        return value
+
+
+def _is_sensitive_environment_name(name: str) -> bool:
+    normalized = _normalize_name(name)
+    return normalized in _SENSITIVE_ENVIRONMENT_NAMES or normalized.endswith(_SENSITIVE_ENVIRONMENT_SUFFIXES)
+
+
+def _is_sensitive_field_name(name: str) -> bool:
+    normalized = _normalize_name(name)
+    return normalized in _SENSITIVE_FIELD_NAMES or normalized.endswith(_SENSITIVE_FIELD_SUFFIXES)
+
+
+def _normalize_name(value: str) -> str:
+    return re.sub(r"[^a-z0-9]+", "_", value.casefold()).strip("_")
+
+
+def _secret_variants(value: str) -> set[str]:
+    variants = {value}
+    stripped = value.strip()
+    if stripped:
+        variants.add(stripped)
+    lowered = stripped.casefold()
+    for prefix in ("basic ", "bearer "):
+        if lowered.startswith(prefix):
+            variants.add(stripped[len(prefix) :])
+    variants.update(_url_secret_variants(stripped))
+    for item in tuple(variants):
+        if item:
+            variants.add(quote(item, safe=""))
+    return {item for item in variants if item}
+
+
+def _url_secret_variants(value: str) -> set[str]:
+    try:
+        parsed = urlsplit(value)
+    except ValueError:
+        return set()
+    if not parsed.netloc:
+        return set()
+    variants = set()
+    if parsed.password:
+        variants.update((parsed.password, unquote(parsed.password)))
+    for name, item in parse_qsl(parsed.query, keep_blank_values=True):
+        if item and _is_sensitive_field_name(name):
+            variants.update((item, unquote(item)))
+    return variants

e2e/bub/src/powercontext_e2e/runner.py

@@ -36,6 +36,7 @@
     ScenarioSpec,
     SessionObservation,
 )
+from .redaction import EvidenceRedactor
 
 Mode = Literal["acceptance", "live", "offline-rescore"]
 Report = EvaluationReport[ScenarioSpec, ReplayObservation, dict[str, str]]
@@ -391,15 +392,17 @@ def _commit() -> str:
 
 
 def _redact(value: str) -> str:
-    secret = os.getenv("BUB_API_KEY")
-    return value.replace(secret, "[REDACTED]") if secret else value
+    return EvidenceRedactor.from_environment().redact_text(value)
 
 
 def write_artifacts(observation: ReplayObservation, report: Report, output_dir: Path) -> None:
+    redactor = EvidenceRedactor.from_environment()
     output_dir.mkdir(parents=True, exist_ok=True)
-    (output_dir / "replay.json").write_text(
-        observation.model_dump_json(by_alias=True, indent=2) + "\n",
-        encoding="utf-8",
+    replay_payload = json.loads(observation.model_dump_json(by_alias=True))
+    _write_evidence(
+        output_dir / "replay.json",
+        json.dumps(redactor.redact(replay_payload), indent=2, ensure_ascii=False) + "\n",
+        redactor,
     )
 
     cases = [
@@ -428,17 +431,23 @@ def write_artifacts(observation: ReplayObservation, report: Report, output_dir:
         "cases": cases,
         "failures": [{"name": failure.name, "error": failure.error_message} for failure in report.failures],
     }
-    (output_dir / "eval-report.json").write_text(
-        json.dumps(report_payload, indent=2, sort_keys=True) + "\n",
-        encoding="utf-8",
+    _write_evidence(
+        output_dir / "eval-report.json",
+        json.dumps(redactor.redact(report_payload), indent=2, sort_keys=True, ensure_ascii=False) + "\n",
+        redactor,
     )
-    (output_dir / "report.md").write_text(
+    _write_evidence(
+        output_dir / "report.md",
         "# PowerContext session replay\n\n"
         f"- Scenario: `{observation.scenario.id}`\n"
         f"- Mode: `{observation.environment.mode}`\n"
         f"- Database: `{observation.environment.database}`\n"
         f"- Status: `{observation.status}`\n\n"
         "## Evaluation\n\n"
         f"```text\n{report.render(include_reasons=True)}\n```\n",
-        encoding="utf-8",
+        redactor,
     )
+
+
+def _write_evidence(path: Path, content: str, redactor: EvidenceRedactor) -> None:
+    path.write_text(redactor.redact_text(content), encoding="utf-8")