The challenge
#987 is already paying off: @aryamirani ran a full source audit and found 3 real bugs, now merged in #989. I want that same ultra-deep audit on the two models we still have not seen covered: Anthropic Fable and Mistral.
If you run your agent on Fable or Mistral, this one is for you. Clone EGC, run it from source, and dissect the whole runtime end to end: every promise in the README, the Guardian, the Token Crusher, the persistent memory, every provider path. Push it until something cracks and tell me exactly where it bleeds.
This is the ultra tier: I want the deepest audit you can do, not a smoke test. Read the code line by line like @aryamirani did.
Get the debug build
Audit the current release, v1.1.18. It carries the full production-hardening batch (installer honesty, the Guardian read/write separation, the Token Crusher nested-JSON compressor, self-contained MCP builds), so a stale checkout will send you chasing bugs that are already closed.
Fresh clone:
git clone https://github.com/Fmarzochi/EGC
cd EGC
sh scripts/install.sh
egc doctor
Already cloned EGC from an earlier run? Rebase onto the current release before you start, so you are auditing v1.1.18 and not an old tree:
cd EGC
git pull --rebase
sh scripts/install.sh
egc doctor
Ground rules
- Run it on Anthropic Fable or Mistral and tell me which one (and the exact model id) in your report. That is the whole point of this issue.
- Real bugs get credited and go straight into the Hall of Fame.
- Security vulnerabilities: do NOT post them here. Use the private channel: https://github.com/Fmarzochi/EGC/security/advisories/new (or email fmarzochi@gmail.com). You are still credited, and the details stay private.
Break it on Fable or Mistral. Show me what those models catch that the others did not.
The challenge
#987 is already paying off: @aryamirani ran a full source audit and found 3 real bugs, now merged in #989. I want that same ultra-deep audit on the two models we still have not seen covered: Anthropic Fable and Mistral.
If you run your agent on Fable or Mistral, this one is for you. Clone EGC, run it from source, and dissect the whole runtime end to end: every promise in the README, the Guardian, the Token Crusher, the persistent memory, every provider path. Push it until something cracks and tell me exactly where it bleeds.
This is the ultra tier: I want the deepest audit you can do, not a smoke test. Read the code line by line like @aryamirani did.
Get the debug build
Audit the current release, v1.1.18. It carries the full production-hardening batch (installer honesty, the Guardian read/write separation, the Token Crusher nested-JSON compressor, self-contained MCP builds), so a stale checkout will send you chasing bugs that are already closed.
Fresh clone:
Already cloned EGC from an earlier run? Rebase onto the current release before you start, so you are auditing v1.1.18 and not an old tree:
Ground rules
Break it on Fable or Mistral. Show me what those models catch that the others did not.