The GitHub Actions job "tests" on airflow-steward.git/triage-prelim-checks has 
failed.
Run started by GitHub user potiuk (triggered by potiuk).

Head commit for run:
32205f6f452d9a67580842efa67b58e8d52f198c / Jarek Potiuk <[email protected]>
triage: mandatory Security Model citation + closed-tracker precedent search

Adds two pre-classification steps to security-issue-triage:

Step 2.5 — Apply the Security Model verbatim. The classifier must quote
the relevant 2-3 sentences of the project's Security Model and explain
how the tracker maps to (or escapes) that wording. A trust-boundary
cheat-sheet codifies the common attacker/target combinations into
default dispositions, so the classifier cannot skip the boundary
analysis silently.

Step 2.6 — Closed-tracker precedent search. Extends the existing
fuzzy-dup search (Step 2a in security-issue-import) to look for closed
'invalid' / 'not CVE worthy' / 'duplicate' trackers — and CVE-allocated
positive precedents — for the same fuzzy keys. A STRONG rejection
precedent lowers proposal confidence and often swings the disposition
from VALID to NOT-CVE-WORTHY.

Motivation: a 2026-05-14 triage sweep against airflow-s/airflow-s got
8 of 9 trackers' first-pass classification wrong — proposed VALID for
cases that had direct NOT-CVE-WORTHY precedents (e.g. airflow-s#258
for the __classname__→import_string pattern, airflow-s#215/#237 for
cross-DAG-read via Execution API). The misclassifications were only
caught after the human triager pushed back twice asking the assistant
to read the Security Model deeply and check past similar issues. This
PR codifies both as mandatory pre-classification steps.

Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>

Report URL: https://github.com/apache/airflow-steward/actions/runs/25854980093

With regards,
GitHub Actions via GitBox


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to