Codex/aura enterprise maturity f - #11
Open
youngbryan97 wants to merge 5 commits into
Open
Conversation
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
There was a problem hiding this comment.
Your trial has ended. Reactivate Greptile to resume code reviews.
PR Summary by QodoPass F maturity: auto action expectations, durable receipts, and effect verification
AI Description
Diagram
High-Level Assessment
Files changed (14)
|
Code Review by Qodo
Context used✅ Compliance rules (platform):
9 rules 1. _action_expectation_digest() swallows exceptions
|
![qodo-code-review[bot]](https://avatars.githubusercontent.com/in/484649?s=60&v=4){kind=small}
Comment on lines
+4508
to
+4517
| def _action_expectation_digest(payload: dict[str, Any]) -> str: | ||
| try: | ||
| encoded = json.dumps( | ||
| payload, | ||
| sort_keys=True, | ||
| separators=(",", ":"), | ||
| default=str, | ||
| ).encode("utf-8") | ||
| except (TypeError, ValueError): | ||
| encoded = str(payload).encode("utf-8", errors="replace") |
There was a problem hiding this comment.
1. _action_expectation_digest() swallows exceptions 📘 Rule violation ☼ Reliability
_action_expectation_digest() catches TypeError/ValueError and falls back to str(payload) without recording a degradation event. This can hide serialization failures and reduce observability of malformed receipt/verdict payloads.
Agent Prompt
## Issue description
`core/capability_engine.py::_action_expectation_digest()` catches `TypeError`/`ValueError` from `json.dumps(...)` but does not call `record_degradation(...)` (or the local wrapper `_record_capability_degradation(...)`). This silently swallows an error condition and violates the requirement that exception handlers must record degradations.
## Issue Context
There is already a project-standard helper `_record_capability_degradation(...)` in this module that wraps `record_degradation(...)`. Use it in the exception handler and capture the exception via `as exc`.
## Fix Focus Areas
- core/capability_engine.py[4508-4517]
ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools
Comment on lines
+314
to
+328
| source_effect = await self._file_effect(full_path) | ||
| dest_effect = await self._file_effect(full_dest) | ||
| effect_verified = bool(dest_effect.get("effect_verified")) | ||
| if source_effect.get("sha256") and dest_effect.get("sha256"): | ||
| effect_verified = source_effect["sha256"] == dest_effect["sha256"] | ||
| return { | ||
| "ok": effect_verified, | ||
| "summary": f"Copied {path} to {dest_path}", | ||
| "path": path, | ||
| "destination": dest_path, | ||
| "source_sha256": source_effect.get("sha256", ""), | ||
| "criteria_results": {"path copied": effect_verified}, | ||
| **dest_effect, | ||
| "effect_verified": effect_verified, | ||
| } |
There was a problem hiding this comment.
2. Copy verifies wrong target 🐞 Bug ≡ Correctness
FileOperationSkill.copy and FileOperationSkill.move verify post-action effects using the requested destination path (full_dest) instead of the final destination returned by the governed operation (result['destination']). When the destination is a directory, this validates the directory rather than the copied/moved file, which can skip sha256-based comparison, lose evidence, and cause expectation checks to be bypassed or downgraded.
Agent Prompt
## Issue description
`file_operation` copy and move compute post-action evidence against `full_dest` (the requested destination), but the governed backend returns a *final* destination path in `result["destination"]` that can differ when the destination is a directory (e.g., `shutil.copy2(src, dst_dir)` returns `<dst_dir>/<basename>` and `shutil.move(src, dst_dir)` similarly resolves to the moved file path). Verifying `full_dest` can therefore verify the wrong object (a directory instead of the copied/moved file), omit `sha256`, skip hash comparison, and cause action expectation checks to be bypassed or downgraded.
## Issue Context
- `ActionExecutor` returns `{"destination": final}` for `op == "copy"` and `op == "move"`.
- `file_write_gateway.copy_path_async` returns `str(shutil.copy2(...))` or `str(shutil.copytree(...))`.
- `file_write_gateway.move_path_async` returns `str(shutil.move(src, dst))`, which changes when `dst` is a directory.
- `_file_effect` only emits `sha256` when the verified path is a file, so verifying a directory destination loses required hash evidence.
## Fix Focus Areas
- core/skills/file_operation.py[266-328]
- core/runtime/action_executor.py[133-153]
- core/runtime/file_write_gateway.py[207-262]
ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools
Comment on lines
+4429
to
+4430
| "move": ("path moved", ["path", "destination", "sha256", "effect_verified"]), | ||
| "copy": ("path copied", ["path", "destination", "sha256", "effect_verified"]), |
There was a problem hiding this comment.
3. Dir move/copy needs sha256 🐞 Bug ≡ Correctness
CapabilityEngine auto-generates ActionExpectations for file_operation move/copy that always require sha256, but file_operation explicitly supports directory paths and _file_effect does not emit sha256 for directories, so successful directory move/copy operations will be downgraded to success_unverified (making ok=False).
Agent Prompt
### Issue description
Auto-generated expectations for `file_operation` move/copy currently require `sha256`, but directory operations are supported and cannot satisfy that evidence key (since `_file_effect` only hashes files). This causes directory move/copy to be treated as missing evidence and downgraded to `success_unverified`.
### Issue Context
- `FileOpInput.path` is documented as "Target file or directory path".
- `_file_effect` only adds `sha256` when `is_file` is true.
- Missing evidence (without missing criteria) maps to `SkillStatus.SUCCESS_UNVERIFIED`.
### Fix Focus Areas
- core/capability_engine.py[4421-4431]
- core/skills/file_operation.py[15-22]
- core/skills/file_operation.py[71-92]
- core/runtime/skill_contract.py[159-188]
### Proposed fix options
Option A (lowest risk):
- Remove `sha256` from the default `move`/`copy` expectation required_evidence list (keep it for write/append/patch).
Option B (stronger verification):
- Extend `file_operation` to emit a deterministic directory digest (e.g., `tree_sha256`) for directories, and update the default expectation logic to require the appropriate digest for directories.
Option C (more invasive):
- Extend `ActionExpectation` to support alternative evidence keys (e.g., `sha256 OR tree_sha256`) so directories and files can both satisfy the contract cleanly.
ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools
Comment on lines
+101
to
+128
| def _file_sha256(path: Path) -> str: | ||
| digest = hashlib.sha256() | ||
| with open(path, "rb") as fh: | ||
| for chunk in iter(lambda: fh.read(1024 * 1024), b""): | ||
| digest.update(chunk) | ||
| return digest.hexdigest() | ||
|
|
||
| @classmethod | ||
| def _core_memory_effect( | ||
| cls, | ||
| block_path: Path, | ||
| *, | ||
| expected_sha256: str, | ||
| ) -> dict[str, Any]: | ||
| exists = block_path.exists() | ||
| effect: dict[str, Any] = { | ||
| "path": str(block_path), | ||
| "exists": exists, | ||
| "effect_verified": False, | ||
| } | ||
| if exists: | ||
| sha256 = cls._file_sha256(block_path) | ||
| effect.update({ | ||
| "sha256": sha256, | ||
| "bytes": block_path.stat().st_size, | ||
| "expected_sha256": expected_sha256, | ||
| "effect_verified": sha256 == expected_sha256, | ||
| }) |
There was a problem hiding this comment.
4. Memory hashing blocks loop 🐞 Bug ➹ Performance
MemoryOpsSkill now reads and hashes the core memory file synchronously inside an async execution path (_core_memory_effect), which can block the event loop as memory blocks grow and increase tail latency for concurrent requests.
Agent Prompt
### Issue description
`MemoryOpsSkill._core_memory_effect` performs blocking file I/O (`open`, `read`, `stat`) and hashing on the event loop thread, but it is called from async `execute` paths. This can block the loop for large core-memory blocks.
### Issue Context
`FileOperationSkill` uses `asyncio.to_thread(...)` for similar filesystem hashing and stat calls, indicating the intended pattern.
### Fix Focus Areas
- core/skills/memory_ops.py[100-171]
- core/skills/file_operation.py[71-88]
### Proposed fix
- Make `_core_memory_effect` async and move the hashing/stat work into `asyncio.to_thread`.
- e.g., `effect = await asyncio.to_thread(self._core_memory_effect_sync, block_path, expected_sha256)`
- Consider also moving the initial `open(block_path).read()` in `_execute_core_memory` to `asyncio.to_thread` for consistency.
ⓘ Copy this prompt and use it to remediate the issue with your preferred AI generation tools
TheITVeteran
pushed a commit
to TheITVeteran/aura
that referenced
this pull request
Jul 17, 2026
The critique's item youngbryan97#11: "a godlike system that believes itself too easily is not godlike. It is brittle." Aura had plan critique (critic_engine) and a project-level claim matrix, but no runtime epistemic critic interrogating a specific claim before it's trusted/emitted. core/cognition/adversarial_audit.py runs the doc's checklist as concrete, grounded checks: overclaiming absolute/grandiose language action_done claim asserts an action but nothing confirms it ran receipt_exists asserted action with no verifiable receipt (checked against the Will audit trail / Outcome Ledger — real grounding, not string matching) stale_memory leans on a memory past its freshness horizon world_state_current leans on a stale world-state snapshot evidence factual assertion with no evidence cited persona_leak first-person phenomenal claims stated as literal fact user_projection asserts the user's state when the live other-agent estimate is low-confidence falsifiability a strong claim with no stated way it could be wrong Returns an AuditReport: per-check findings, a severity-weighted risk score, a verdict (trust/caveat/block), and concrete suggested caveats — turning "honest assessments over validation" into a mechanism. Cross-wired to the receipt substrate and the other-agent estimate so the checks reflect real runtime state. Registered as 'adversarial_auditor' (lazy, required=False). Also strengthened OtherAgentStateEstimator.overall_confidence: average over channels that actually have evidence rather than all channels (averaging in never-observed channels structurally capped a strong, focused read too low) — a more honest confidence used by the audit's projection grounding, the value model's protect-future-self check, and the ladder's social signals. 21 social tests still green. tests/test_adversarial_audit.py: 16 tests (every check, receipt verified via Will, projection grounded/ungrounded against the estimate, severity-weighted risk, block aggregation). 95 green across the full new social/world-model/memory/values/cognition/agency cluster. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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.
No description provided.