Occurrence: #119 (comment)
The coordinator audits every Epic record before posting it with a small Python check (non-ASCII count must be 0; no closing keyword followed by an issue reference) that ends with sys.exit(1) on a hit. In this occurrence the audit reported a hit (the word closed followed by a colon and the reference to issue 120, in the sentence listing the sixteen finished sub-issues) but the gh api -X POST that followed was written on its own line after the heredoc instead of being chained with &&, so the post ran regardless of the audit's exit status. The comment is an issue comment, where closing keywords have no effect, and all cited issues were already closed, so nothing changed on GitHub; the record is otherwise accurate and stays unedited. The failure class is a fail-closed check whose result is not wired to the action it guards. The plan-management and session-orchestration skills describe the audit intent but do not require the post step to be conditional on the audit, so any agent scripting the same read-audit-post sequence can repeat it.
This issue records the failure class only (first occurrence); it does not authorize skill or script changes.
Occurrence: #119 (comment)
The coordinator audits every Epic record before posting it with a small Python check (non-ASCII count must be 0; no closing keyword followed by an issue reference) that ends with
sys.exit(1)on a hit. In this occurrence the audit reported a hit (the wordclosedfollowed by a colon and the reference to issue 120, in the sentence listing the sixteen finished sub-issues) but thegh api -X POSTthat followed was written on its own line after the heredoc instead of being chained with&&, so the post ran regardless of the audit's exit status. The comment is an issue comment, where closing keywords have no effect, and all cited issues were already closed, so nothing changed on GitHub; the record is otherwise accurate and stays unedited. The failure class is a fail-closed check whose result is not wired to the action it guards. The plan-management and session-orchestration skills describe the audit intent but do not require the post step to be conditional on the audit, so any agent scripting the same read-audit-post sequence can repeat it.This issue records the failure class only (first occurrence); it does not authorize skill or script changes.