Adam Moussa ea1845d41d fix(security): fence and neutralize untrusted Slack channel description in agent prompt ... Hardens SLACK-PI-001 (sh-security-review). Slack channel topic/purpose is editable by ordinary channel members and flowed verbatim into the agent LLM prompt behind only a prose 'untrusted' label — an indirect prompt-injection vector for an agent with network egress and repo write. Now strip leading markdown structural tokens per line (so it can't forge the prompt's real request/section delimiters), cap length, and wrap it in a per-render unguessable sentinel fence (so injected text can't spoof a closing marker to escape the data block). Deliberately diverges from upstream #1633.		2026-07-03 16:08:35 -04:00
..
__init__.py	refactor: adopt modular webhook architecture (#1621) + port fork customizations (#85)	2026-06-30 18:46:46 -04:00
github.py	refactor: adopt modular webhook architecture (#1621) + port fork customizations (#85)	2026-06-30 18:46:46 -04:00
linear.py	refactor: adopt modular webhook architecture (#1621) + port fork customizations (#85)	2026-06-30 18:46:46 -04:00
slack.py	fix(security): fence and neutralize untrusted Slack channel description in agent prompt	2026-07-03 16:08:35 -04:00