52 lines
1.9 KiB
Markdown
52 lines
1.9 KiB
Markdown
# Source Review - AINative_OKR_Claude_GHCP
|
|
|
|
## Execution Result
|
|
|
|
After enabling executable bits for Bash scripts, the main Bash harness test passed:
|
|
|
|
```bash
|
|
cd AINative_OKR_Claude_GHCP
|
|
chmod +x .specify/scripts/bash/*.sh .specify/tests/run-casan4-harness-tests.sh
|
|
bash .specify/tests/run-casan4-harness-tests.sh
|
|
```
|
|
|
|
Result: all built-in checks passed.
|
|
|
|
PowerShell test was not executed because the current machine does not have `pwsh` or `powershell`.
|
|
|
|
## Issues Found
|
|
|
|
| Severity | Issue | Impact |
|
|
|---|---|---|
|
|
| P0 | `.sh` files were not executable in the source folder | Test failed before chmod; package was not self-verifying on macOS/Linux. |
|
|
| P0 | Private key material was not blocked by Bash H4 due `grep` option parsing | `-----BEGIN PRIVATE KEY-----` passed through as safe content. |
|
|
| P1 | Idempotency cache returned before H4/H5/H6 gates | Cache hit skipped fresh security/governance/metrics evidence. |
|
|
| P1 | PowerShell path was not locally verifiable | Good Windows evidence path, but not proven in this environment. |
|
|
| P2 | Several Level 5 controls were policy/config only | Useful direction, but not enough for full production Level 5. |
|
|
|
|
## Source Score
|
|
|
|
Executable-adjusted score: approximately 85.1 / 100.
|
|
|
|
Raw submission score before chmod: lower, because the first automated test failed on permissions.
|
|
|
|
## What Was Kept
|
|
|
|
- Self-security protocol.
|
|
- Governance risk policy.
|
|
- Hallucination prevention checklist.
|
|
- Tool registry policy.
|
|
- Rate-limit and hallucination tracking configs.
|
|
- Expanded tool registry metadata.
|
|
- PowerShell harness scripts for Windows review.
|
|
- Compliance and dashboard generation ideas.
|
|
|
|
## What Was Fixed in the Refined Package
|
|
|
|
- Preserved executable Bash baseline from CASAN4.
|
|
- Fixed private key detection in Bash H4.
|
|
- Fixed cache-hit evidence bypass in Bash wrapper.
|
|
- Added tool-call audit log from registry and command wrapper.
|
|
- Added regression tests for jailbreak, private key, cache trace, and tool audit.
|
|
|