Conversation
The 2026-04-23 blocked-task escalation on eigendark-website#77 cascaded through three distinct bugs; this change fixes all three. 1. Agents (especially codex) sometimes copied the .agent_result.md prompt template verbatim — including the literal instruction text "One line. Required when STATUS is partial or blocked..." as the BLOCKER_CODE value and "- bullet\n- bullet" as DONE. The prose placeholders looked plausibly like content. Replace with unambiguous `<...>` markers and add an explicit "never copy the <...> placeholders" rule. 2. The escalation snapshot (`_extract_task_result_snapshot` in github_dispatcher) read the raw log text after `BLOCKER_CODE:` without validating it against `VALID_BLOCKER_CODES`. That is why the Telegram escalation displayed the entire echoed instruction prose as the blocker code. New `_sanitize_snapshot_blocker_code` maps anything unknown to `invalid_result_contract` so error-pattern aggregation stays useful. 3. Agent PRs opened via `_create_prs_for_orphan_branches` (pr_monitor) and via github_sync used "Automated changes for issue #N" as the PR body. That is NOT a GitHub closing keyword, so merging the PR left its linked issue open with the `in-progress` label, which kept the blocked-task poller firing on long-completed work. Both paths now lead with `Closes #N`. Together these ensure a prompt-echo incident (a) produces a crisp `invalid_result_contract` signature the operator can act on, and (b) cannot cascade into a day-long escalation on a PR that already merged.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
The 2026-04-23 blocked-task escalation on eigendark-website#77 was caused by three cascading bugs. Fixing all three.
Context
Telegram alert:
The blocker "code" was the literal prompt template text. Root causes:
1. Prose placeholders in the .agent_result.md template
Agents (codex in this case) copied the template verbatim. The old template had "One line. Required when STATUS..." and "- bullet / - bullet" which look plausibly like content.
Fix: replace with `<...>` markers, explicit "never copy the placeholders" rule.
2. Snapshot extractor didn't validate BLOCKER_CODE
`_extract_task_result_snapshot` read the raw log text after `BLOCKER_CODE:` without checking `VALID_BLOCKER_CODES`. That's why the echoed prose flowed into the Telegram and into the error-pattern aggregator.
Fix: new `_sanitize_snapshot_blocker_code` maps unknown values to `invalid_result_contract` so aggregation stays useful.
3. Agent PRs didn't auto-close their linked issue
Both `_create_prs_for_orphan_branches` (pr_monitor) and github_sync wrote "Automated changes for issue #N" as the PR body — not a GitHub closing keyword. Merging the PR left the issue open with `in-progress`, which kept the blocked-task poller firing on work that was long done.
Fix: both paths now lead with `Closes #N`.
Also
Closed eigendark-website#77 manually (the page has been rendering correctly since PR #88 merged yesterday).
Test plan