feat: per-turn hook audit log (#280 mitigation 3) - #314
Merged
Conversation
Adds hook_audit.jsonl alongside the existing hook_telemetry.jsonl. Where telemetry records counts, audit records the *payload* — the exact rendered <aelfrice-memory> / <aelfrice-baseline> block injected into a given turn — so a reviewer can later see what the hook actually emitted on the turn a suspect critique landed. Defaults match the spec in docs/hook_hardening.md: - Default-on. Opt-out via AELFRICE_HOOK_AUDIT=0 env var or [hook_audit] enabled = false in .aelfrice.toml. - 10 MB cap, configurable via [hook_audit] max_bytes. - Single-slot rotation: live -> hook_audit.jsonl.1 on rollover. - Fail-soft: any I/O error is traced to stderr and swallowed. Wires into both user_prompt_submit (records emitted block + locked count + n_beliefs + session_id when available) and session_start (records baseline block, hook='session_start'). The session-start path now also reads session_id from the payload (best-effort) and exposes _retrieve_baseline_with_block() as the non-formatting-only retrieval helper so the hook can pass hits to the audit record without re-querying the store. Closes the audit-log half of #280; mitigations 1+2 (framing-tag contract + render-time belief-content escape) already shipped via PR #292.
|
Warning Rate limit exceeded
To keep reviews running without waiting, you can enable usage-based add-on for your organization. This allows additional reviews beyond the hourly cap. Account admins can enable it under billing. ⌛ How to resolve this issue?After the wait time has elapsed, a review can be triggered using the We recommend that you space out your commits to avoid hitting the rate limit. 🚦 How do rate limits work?CodeRabbit enforces hourly rate limits for each developer per organization. Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout. Please see our FAQ for further information. ℹ️ Review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Run ID: 📒 Files selected for processing (2)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Review rate limit: 0/1 reviews remaining, refill in 36 minutes and 6 seconds.Comment |
Reviewer's GuideImplements a per-turn hook audit logging system that records the exact rendered hook block for user_prompt_submit and session_start into a new JSONL audit file with configurable enablement and rotation, plus supporting config resolution, session_id extraction, and comprehensive tests. Sequence diagram for per-turn hook audit write on user_prompt_submitsequenceDiagram
actor Harness
participant Hook as user_prompt_submit
participant Extract as _extract_session_id
participant Audit as _write_hook_audit_record
participant Config as load_hook_audit_config
participant DBPath as db_path
participant FS as Filesystem
Harness->>Hook: invoke with stdin payload
Hook->>Hook: _extract_prompt(raw)
Hook->>Extract: _extract_session_id(raw)
Extract-->>Hook: session_id or None
Hook->>Hook: retrieve(..., token_budget)
Hook-->>Harness: write rendered_block to stdout
Hook->>Audit: _write_hook_audit_record(hook, prompt, rendered_block, n_beliefs, n_locked, session_id)
activate Audit
Audit->>Config: load_hook_audit_config()
Config-->>Audit: HookAuditConfig(enabled, max_bytes)
Audit->>Audit: check enabled
alt audit disabled
Audit-->>Hook: return (no-op)
else audit enabled
Audit->>DBPath: db_path()
DBPath-->>Audit: Path to memory.db
Audit->>Audit: _audit_path_for_db(db_path)
Audit->>FS: _append_audit(audit_path, record, max_bytes)
FS-->>Audit: write/rotate or error
Audit-->>Hook: return
end
deactivate Audit
Sequence diagram for per-turn hook audit write on session_startsequenceDiagram
actor Harness
participant Hook as session_start
participant Extract as _extract_session_id
participant Retrieve as _retrieve_baseline_with_block
participant Audit as _write_hook_audit_record
participant Config as load_hook_audit_config
participant DBPath as db_path
participant FS as Filesystem
Harness->>Hook: invoke with stdin payload
Hook->>Hook: read raw from stdin
Hook->>Extract: _extract_session_id(raw)
Extract-->>Hook: session_id or None
Hook->>Retrieve: _retrieve_baseline_with_block(token_budget)
Retrieve->>Retrieve: retrieve(store, "", token_budget)
Retrieve-->>Hook: hits, rendered_block
alt rendered_block empty
Hook-->>Harness: return 0 (no baseline block)
else rendered_block non-empty
Hook-->>Harness: write rendered_block to stdout
Hook->>Audit: _write_hook_audit_record(hook, prompt="", rendered_block, n_beliefs=len(hits), n_locked, session_id)
activate Audit
Audit->>Config: load_hook_audit_config()
Config-->>Audit: HookAuditConfig(enabled, max_bytes)
Audit->>Audit: check enabled
alt audit disabled
Audit-->>Hook: return (no-op)
else audit enabled
Audit->>DBPath: db_path()
DBPath-->>Audit: Path to memory.db
Audit->>Audit: _audit_path_for_db(db_path)
Audit->>FS: _append_audit(audit_path, record, max_bytes)
FS-->>Audit: write/rotate or error
Audit-->>Hook: return
end
deactivate Audit
end
Class diagram for new hook audit structures and helpersclassDiagram
class HookAuditConfig {
<<dataclass>>
bool enabled
int max_bytes
}
class HookAuditModule {
+int AUDIT_DEFAULT_MAX_BYTES
+int AUDIT_PROMPT_PREFIX_CAP
+str AUDIT_FILENAME
+str AUDIT_ROTATED_SUFFIX
+str AUDIT_HOOK_USER_PROMPT_SUBMIT
+str AUDIT_HOOK_SESSION_START
+HookAuditConfig load_hook_audit_config(start, env, stderr)
+Path _audit_path_for_db(db_path_val)
+void _append_audit(audit_path, record, max_bytes, stderr)
+void _write_hook_audit_record(hook, prompt, rendered_block, n_beliefs, n_locked, session_id, config, stderr)
+list~dict~ read_hook_audit(path)
+str _extract_session_id(raw)
+tuple~list~Belief~~str~ _retrieve_baseline_with_block(token_budget)
}
class Belief {
}
HookAuditModule --> HookAuditConfig : uses
HookAuditModule --> Belief : returns list of
class HookHooks {
+int user_prompt_submit(token_budget, sin, sout, serr)
+int session_start(token_budget, sin, sout, serr)
}
HookHooks --> HookAuditModule : calls _write_hook_audit_record
HookHooks --> HookAuditModule : calls _extract_session_id
HookHooks --> HookAuditModule : calls _retrieve_baseline_with_block
File-Level Changes
Possibly linked issues
Tips and commandsInteracting with Sourcery
Customizing Your ExperienceAccess your dashboard to:
Getting Help
|
![sourcery-ai[bot]](https://avatars.githubusercontent.com/in/48477?s=60&v=4){kind=small}
There was a problem hiding this comment.
Hey - I've found 3 issues, and left some high level feedback:
- Both
user_prompt_submitandsession_startnow parse the JSON payload separately in_extract_promptand_extract_session_id; consider refactoring to parse once and thread the decoded payload through to avoid redundant work and potential divergence in behavior. _append_auditcallsos.fsyncon every write, which can be quite expensive on some filesystems; if this log is not strictly durability-critical, consider making fsync configurable or relaxing it (e.g., periodic or best-effort) to reduce latency in hot paths.- The
test_audit_write_failsoft_on_unwriteable_pathtest is annotated in comments as POSIX-only but is not conditionally skipped on Windows; consider adding apytest.mark.skipifor equivalent guard so this test doesn’t become flaky or fail on non-POSIX platforms.
Prompt for AI Agents
Please address the comments from this code review:
## Overall Comments
- Both `user_prompt_submit` and `session_start` now parse the JSON payload separately in `_extract_prompt` and `_extract_session_id`; consider refactoring to parse once and thread the decoded payload through to avoid redundant work and potential divergence in behavior.
- `_append_audit` calls `os.fsync` on every write, which can be quite expensive on some filesystems; if this log is not strictly durability-critical, consider making fsync configurable or relaxing it (e.g., periodic or best-effort) to reduce latency in hot paths.
- The `test_audit_write_failsoft_on_unwriteable_path` test is annotated in comments as POSIX-only but is not conditionally skipped on Windows; consider adding a `pytest.mark.skipif` or equivalent guard so this test doesn’t become flaky or fail on non-POSIX platforms.
## Individual Comments
### Comment 1
<location path="src/aelfrice/hook.py" line_range="432-435" />
<code_context>
+ try:
+ audit_path.parent.mkdir(parents=True, exist_ok=True)
+ line = json.dumps(record) + "\n"
+ with open(audit_path, "a", encoding="utf-8") as f:
+ f.write(line)
+ f.flush()
+ os.fsync(f.fileno())
+ if audit_path.stat().st_size > max_bytes:
+ rotated = audit_path.with_name(
</code_context>
<issue_to_address>
**suggestion (performance):** Per-record `fsync` could be expensive under high hook volume.
This gives strong durability but may become a bottleneck on busy or slow filesystems. Consider making the sync policy configurable (e.g., batch/periodic sync or OS buffering only) so operators can tune the durability vs throughput trade-off for their deployment.
Suggested implementation:
```python
with open(audit_path, "a", encoding="utf-8") as f:
f.write(line)
# sync_policy controls durability vs throughput:
# - "always": flush + fsync each record (strong durability, slowest)
# - "flush": flush only, rely on OS to sync (middle ground)
# - "none": rely entirely on OS buffering (fastest)
if sync_policy in ("always", "flush"):
f.flush()
if sync_policy == "always":
os.fsync(f.fileno())
```
To fully implement configurability of the sync behavior, you should also:
1. Update the function signature that contains this code (likely a hook audit/write helper) to accept a new parameter, for example:
- `sync_policy: str = "always"`
2. Extend the function docstring to document the `sync_policy` parameter and the supported values: `"always"`, `"flush"`, and `"none"`, and the durability/throughput trade-offs.
3. Propagate the new parameter from configuration:
- Add a configuration option (CLI flag, config file value, or environment variable) that determines the desired `sync_policy`.
- Thread that configuration value through to all call sites of this function.
4. Ensure that any direct calls to this function in tests or other modules are updated to either:
- Rely on the default `sync_policy="always"` for current behavior, or
- Explicitly pass `"flush"` or `"none"` where stronger throughput is desired.
5. Optionally, validate `sync_policy` inside the function (e.g., raise `ValueError` or fall back to `"always"` if an unknown value is provided) to avoid silent misconfiguration.
</issue_to_address>
### Comment 2
<location path="tests/test_hook_audit.py" line_range="148-157" />
<code_context>
+def test_user_prompt_submit_writes_audit_record(
</code_context>
<issue_to_address>
**suggestion (testing):** Add focused tests for `_extract_session_id` edge cases (invalid JSON, non-dict payload, missing/empty `session_id`).
Current integration tests only cover the happy path; they don’t exercise malformed JSON, non-dict roots, or missing/invalid `session_id` values. Direct unit tests on `_extract_session_id` for these cases would better guard the audit trail behavior and future payload shape changes.
Suggested implementation:
```python
# ---------------------------------------------------------------------------
# _extract_session_id unit tests
# ---------------------------------------------------------------------------
def test_extract_session_id_invalid_json() -> None:
# Malformed JSON should not raise and should return no session id
assert _extract_session_id("{not-json") is None
def test_extract_session_id_non_dict_root() -> None:
# Non-object JSON roots should be ignored
assert _extract_session_id('["not", "a", "dict"]') is None
assert _extract_session_id('"plain-string"') is None
assert _extract_session_id("123") is None
assert _extract_session_id("null") is None
def test_extract_session_id_missing_key() -> None:
# Valid JSON object without a session_id key should yield no session id
assert _extract_session_id("{}") is None
assert _extract_session_id('{"other_key": "value"}') is None
def test_extract_session_id_empty_or_non_string_value() -> None:
# Empty or non-string session_id values should be treated as absent
assert _extract_session_id('{"session_id": ""}') is None
assert _extract_session_id('{"session_id": null}') is None
assert _extract_session_id('{"session_id": 123}') is None
assert _extract_session_id('{"session_id": {}}') is None
assert _extract_session_id('{"session_id": []}') is None
# ---------------------------------------------------------------------------
# Hook integration: writes a record on UserPromptSubmit fire
# ---------------------------------------------------------------------------
def test_user_prompt_submit_writes_audit_record(
tmp_path: Path, monkeypatch: pytest.MonkeyPatch,
) -> None:
db = tmp_path / "memory.db"
_seed_db(db, [_mk("F1", "the kitchen is full of bananas")])
_set_db(monkeypatch, db)
monkeypatch.delenv("AELFRICE_HOOK_AUDIT", raising=False)
monkeypatch.chdir(tmp_path)
sin = io.StringIO(_payload("bananas", session_id="sess-abc"))
sout = io.StringIO()
rc = user_prompt_submit(stdin=sin, stdout=sout)
```
1. Ensure `_extract_session_id` is imported into `tests/test_hook_audit.py` from the module under test (likely alongside `user_prompt_submit` and `load_hook_audit_config`). For example:
- `from aelfrice.hook_audit import _extract_session_id`
2. If the actual contract of `_extract_session_id` differs (e.g., it raises on malformed JSON or returns `""` instead of `None`), adjust the assertions accordingly to match the existing implementation.
</issue_to_address>
### Comment 3
<location path="tests/test_hook_audit.py" line_range="79-84" />
<code_context>
+ assert cfg.max_bytes == AUDIT_DEFAULT_MAX_BYTES
+
+
+def test_env_disable_overrides_toml(tmp_path: Path) -> None:
+ (tmp_path / ".aelfrice.toml").write_text(
+ "[hook_audit]\nenabled = true\n", encoding="utf-8",
+ )
+ cfg = load_hook_audit_config(
+ start=tmp_path, env={"AELFRICE_HOOK_AUDIT": "0"},
+ )
+ assert cfg.enabled is False
+
+
</code_context>
<issue_to_address>
**suggestion (testing):** Add a test to confirm that `AELFRICE_HOOK_AUDIT` values other than "0" do *not* disable auditing.
Current tests cover only the disabling case (`AELFRICE_HOOK_AUDIT="0"` and TOML). Since the code checks specifically for `env_val.strip() == "0"`, please add a test where `AELFRICE_HOOK_AUDIT` is set to a non-zero/non-empty value (e.g., "1" or "foo") and verify that auditing stays enabled and config resolution behaves normally. This will lock in the intended env override semantics for future changes.
```suggestion
def test_default_config_is_enabled_with_default_max_bytes(
tmp_path: Path,
) -> None:
cfg = load_hook_audit_config(start=tmp_path, env={})
assert cfg.enabled is True
assert cfg.max_bytes == AUDIT_DEFAULT_MAX_BYTES
def test_env_nonzero_does_not_disable_toml_enabled(tmp_path: Path) -> None:
(tmp_path / ".aelfrice.toml").write_text(
"[hook_audit]\nenabled = true\n",
encoding="utf-8",
)
cfg = load_hook_audit_config(
start=tmp_path,
env={"AELFRICE_HOOK_AUDIT": "1"},
)
assert cfg.enabled is True
```
</issue_to_address>Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.
Comment on lines
+432
to
+435
| with open(audit_path, "a", encoding="utf-8") as f: | ||
| f.write(line) | ||
| f.flush() | ||
| os.fsync(f.fileno()) |
There was a problem hiding this comment.
suggestion (performance): Per-record fsync could be expensive under high hook volume.
This gives strong durability but may become a bottleneck on busy or slow filesystems. Consider making the sync policy configurable (e.g., batch/periodic sync or OS buffering only) so operators can tune the durability vs throughput trade-off for their deployment.
Suggested implementation:
with open(audit_path, "a", encoding="utf-8") as f:
f.write(line)
# sync_policy controls durability vs throughput:
# - "always": flush + fsync each record (strong durability, slowest)
# - "flush": flush only, rely on OS to sync (middle ground)
# - "none": rely entirely on OS buffering (fastest)
if sync_policy in ("always", "flush"):
f.flush()
if sync_policy == "always":
os.fsync(f.fileno())To fully implement configurability of the sync behavior, you should also:
- Update the function signature that contains this code (likely a hook audit/write helper) to accept a new parameter, for example:
sync_policy: str = "always"
- Extend the function docstring to document the
sync_policyparameter and the supported values:"always","flush", and"none", and the durability/throughput trade-offs. - Propagate the new parameter from configuration:
- Add a configuration option (CLI flag, config file value, or environment variable) that determines the desired
sync_policy. - Thread that configuration value through to all call sites of this function.
- Add a configuration option (CLI flag, config file value, or environment variable) that determines the desired
- Ensure that any direct calls to this function in tests or other modules are updated to either:
- Rely on the default
sync_policy="always"for current behavior, or - Explicitly pass
"flush"or"none"where stronger throughput is desired.
- Rely on the default
- Optionally, validate
sync_policyinside the function (e.g., raiseValueErroror fall back to"always"if an unknown value is provided) to avoid silent misconfiguration.
Comment on lines
+148
to
+157
| def test_user_prompt_submit_writes_audit_record( | ||
| tmp_path: Path, monkeypatch: pytest.MonkeyPatch, | ||
| ) -> None: | ||
| db = tmp_path / "memory.db" | ||
| _seed_db(db, [_mk("F1", "the kitchen is full of bananas")]) | ||
| _set_db(monkeypatch, db) | ||
| monkeypatch.delenv("AELFRICE_HOOK_AUDIT", raising=False) | ||
| monkeypatch.chdir(tmp_path) | ||
| sin = io.StringIO(_payload("bananas", session_id="sess-abc")) | ||
| sout = io.StringIO() |
There was a problem hiding this comment.
suggestion (testing): Add focused tests for _extract_session_id edge cases (invalid JSON, non-dict payload, missing/empty session_id).
Current integration tests only cover the happy path; they don’t exercise malformed JSON, non-dict roots, or missing/invalid session_id values. Direct unit tests on _extract_session_id for these cases would better guard the audit trail behavior and future payload shape changes.
Suggested implementation:
# ---------------------------------------------------------------------------
# _extract_session_id unit tests
# ---------------------------------------------------------------------------
def test_extract_session_id_invalid_json() -> None:
# Malformed JSON should not raise and should return no session id
assert _extract_session_id("{not-json") is None
def test_extract_session_id_non_dict_root() -> None:
# Non-object JSON roots should be ignored
assert _extract_session_id('["not", "a", "dict"]') is None
assert _extract_session_id('"plain-string"') is None
assert _extract_session_id("123") is None
assert _extract_session_id("null") is None
def test_extract_session_id_missing_key() -> None:
# Valid JSON object without a session_id key should yield no session id
assert _extract_session_id("{}") is None
assert _extract_session_id('{"other_key": "value"}') is None
def test_extract_session_id_empty_or_non_string_value() -> None:
# Empty or non-string session_id values should be treated as absent
assert _extract_session_id('{"session_id": ""}') is None
assert _extract_session_id('{"session_id": null}') is None
assert _extract_session_id('{"session_id": 123}') is None
assert _extract_session_id('{"session_id": {}}') is None
assert _extract_session_id('{"session_id": []}') is None
# ---------------------------------------------------------------------------
# Hook integration: writes a record on UserPromptSubmit fire
# ---------------------------------------------------------------------------
def test_user_prompt_submit_writes_audit_record(
tmp_path: Path, monkeypatch: pytest.MonkeyPatch,
) -> None:
db = tmp_path / "memory.db"
_seed_db(db, [_mk("F1", "the kitchen is full of bananas")])
_set_db(monkeypatch, db)
monkeypatch.delenv("AELFRICE_HOOK_AUDIT", raising=False)
monkeypatch.chdir(tmp_path)
sin = io.StringIO(_payload("bananas", session_id="sess-abc"))
sout = io.StringIO()
rc = user_prompt_submit(stdin=sin, stdout=sout)- Ensure
_extract_session_idis imported intotests/test_hook_audit.pyfrom the module under test (likely alongsideuser_prompt_submitandload_hook_audit_config). For example:from aelfrice.hook_audit import _extract_session_id
- If the actual contract of
_extract_session_iddiffers (e.g., it raises on malformed JSON or returns""instead ofNone), adjust the assertions accordingly to match the existing implementation.
Comment on lines
+79
to
+84
| def test_default_config_is_enabled_with_default_max_bytes( | ||
| tmp_path: Path, | ||
| ) -> None: | ||
| cfg = load_hook_audit_config(start=tmp_path, env={}) | ||
| assert cfg.enabled is True | ||
| assert cfg.max_bytes == AUDIT_DEFAULT_MAX_BYTES |
There was a problem hiding this comment.
suggestion (testing): Add a test to confirm that AELFRICE_HOOK_AUDIT values other than "0" do not disable auditing.
Current tests cover only the disabling case (AELFRICE_HOOK_AUDIT="0" and TOML). Since the code checks specifically for env_val.strip() == "0", please add a test where AELFRICE_HOOK_AUDIT is set to a non-zero/non-empty value (e.g., "1" or "foo") and verify that auditing stays enabled and config resolution behaves normally. This will lock in the intended env override semantics for future changes.
Suggested change
| def test_default_config_is_enabled_with_default_max_bytes( | |
| tmp_path: Path, | |
| ) -> None: | |
| cfg = load_hook_audit_config(start=tmp_path, env={}) | |
| assert cfg.enabled is True | |
| assert cfg.max_bytes == AUDIT_DEFAULT_MAX_BYTES | |
| def test_default_config_is_enabled_with_default_max_bytes( | |
| tmp_path: Path, | |
| ) -> None: | |
| cfg = load_hook_audit_config(start=tmp_path, env={}) | |
| assert cfg.enabled is True | |
| assert cfg.max_bytes == AUDIT_DEFAULT_MAX_BYTES | |
| def test_env_nonzero_does_not_disable_toml_enabled(tmp_path: Path) -> None: | |
| (tmp_path / ".aelfrice.toml").write_text( | |
| "[hook_audit]\nenabled = true\n", | |
| encoding="utf-8", | |
| ) | |
| cfg = load_hook_audit_config( | |
| start=tmp_path, | |
| env={"AELFRICE_HOOK_AUDIT": "1"}, | |
| ) | |
| assert cfg.enabled is True |
Owner
Author
|
[claim:review:Setr:2026-04-29T16:25:13Z] |
Owner
Author
|
[release:review:Setr:2026-04-29T16:26:26Z] |
This was referenced Apr 30, 2026
This was referenced Apr 30, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds the per-turn audit log called for in
docs/hook_hardening.md(#280 mitigation 3) — a JSONL sibling of the existinghook_telemetry.jsonlthat records the exact rendered hook block on every fire. Telemetry records counts; audit records what was actually injected.<git-common-dir>/aelfrice/hook_audit.jsonl.user_prompt_submitandsession_start.AELFRICE_HOOK_AUDIT=0env or[hook_audit] enabled = falsein.aelfrice.toml.[hook_audit] max_bytes); single-slot rotation tohook_audit.jsonl.1on rollover.Spec references in this repo:
docs/hook_hardening.md— full design.docs/hook_hardening.md#3-per-turn-audit-log— record schema (matches what this PR implements).Mitigations 1 and 2 from the same spec (framing-tag contract + render-time belief-content escape) already shipped via #292; this PR closes the third and final mitigation.
Test plan
tests/test_hook_audit.pycovering: default config, env disable, TOML disable, max_bytes override, malformed/wrong-typed TOML graceful degradation, UserPromptSubmit + SessionStart write paths, n_locked / session_id / prompt_prefix capping, no-op when disabled, no-op when retrieval is empty, rotation at threshold, single-slot rotation overwriting prior.1, read API + corruption handling, fail-soft on unwriteable path.Summary by Sourcery
Introduce a per-turn hook audit log that records rendered hook blocks for user prompt submissions and session starts, with configurable opt-out and size-based rotation.
New Features:
Enhancements:
Tests: