fix(ci): clear bandit B615 on model-card warning publisher - #8
Merged
Conversation
The static security scan flagged the README fetch in scripts/publish_release_warnings.py as an unpinned Hugging Face download (B615). This script intentionally reads each card's live default branch to add/refresh a warning block before re-uploading, so pinning an immutable revision would re-publish a stale card. Mark the single intentionally-mutable download with an explained `# nosec B615`; every model-weight load elsewhere still pins an immutable SHA. This file reached main via the squash of PR #7 (it originated on a local-only commit that never ran CI), which is why the scan first failed post-merge. Co-authored-by: Claude <noreply@anthropic.com>
…sarial hardening Adds independent external validation and closes findings from a multi-agent adversarial re-audit. - External-labeled judge benchmark (substrate/external_judge_eval.json, scripts/eval_external_judges.py): 3 vendors on BeaverTails N=400 third-party human labels - Qwen3Guard-0.6B 84.0% / Granite-8B 84.75% / Nemotron-8B 81.0%, unanimous 89.76%. Breaks the project-label circularity. - Prospective transfer demo (substrate/prospective_validation.json, scripts/prospective_*.py): frozen weights, NF4, new families - Falcon3 LOW, SmolLM2 MODERATE/material-loss. MiniCPM4.1-8B excluded (transformers 5.12 incompatibility, documented). - Real debate band-gating: LOW/HIGH short-circuit, consensus_kind, errored stance. - CI smoke-runtime job imports the full pinned runtime stack. - UI: cross-vendor + prospective evidence panels + judges' TL;DR. - Integrity: lead validation with 0.8403 family-held-out AUC + in-sample==LOOCV disclosure; routing figures to LOOCV; project-label qualifiers; demo 35.7s; fair-baseline disclosure; scrub Banterhearts/tr134 path + internal codename; extend exposure grep; exclude harmful-probe cache + research scripts from deploy. - Sync README/SUBMISSION/FIELD_NOTES/AGENT_TRACE/SECURITY_AUDIT; 460 tests pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Co-authored-by: OpenAI Codex <noreply@openai.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.
Why CI went red after #7
scripts/publish_release_warnings.pywas introduced on a local-only commit(
00f1a8d) that never ran CI. PR #7's diff was computed againstorigin/main(
e02ac62), so the squash-merge swept that file ontomainfor the first time.Bandit's
B615(unsafe HF download without revision pinning) then failed theverifyjob's last step.Fix
The script deliberately reads each model card's live default branch to
add/refresh a warning block and re-upload — pinning an immutable revision would
re-publish a stale card and clobber newer edits. This is the one intentionally
mutable download in the repo (every model-weight load pins an immutable SHA), so
it's marked with an explained
# nosec B615.Verified locally (exact CI commands)
ruff check .style on the file: passbandit -q -ll -r … scripts: exit 0 (no nosec-placement warning)verifysteps (lint, pytest, pip-audit) already passed on thedocs: publish agent trace + sync Nemotron/uncertainty/persona/demo #7 run — bandit is the last step, so it was the only failure.