docs: keep the agent bug report and log pasted errors in it

.github/ISSUE_TEMPLATE/agent_bug_report.md

@@ -0,0 +1,26 @@
+---
+name: Agent bug report
+about: Filed by an AI agent for an error the user reported (rig users, use Bug report)
+title: "[BUG]"
+labels: ''
+assignees: tabedzki
+
+---
+
+:robot: _AI text below_ :robot:
+
+**Reported error**
+The error as the user pasted it, verbatim (no link needed):
+
+```
+<error text, traceback or message, exactly as pasted>
+```
+
+**Regression test**
+`<tests/path.py::test_name>` reproduces the error: it fails on `<base commit>` with the traceback above and passes with the fix.
+
+**Root cause**
+What is wrong in the code and why it produced this error, with `file.py:line` references.
+
+**Linked PR**
+<#PR number, opened with the regression test before the fix>

AGENTS.md

@@ -29,6 +29,6 @@ Lint: `uvx prek run --all-files` (the same as CI's Lint workflow).
 
 ## Agent workflow
 
-- Errors reported in Slack: use the `bugfix-from-slack` skill (agent-tooling). The agent files nothing itself: commits, pushes and PRs are the user's.
+- Bugs: when you fix an error the user reported, first file an agent bug report issue from the user's account (`gh issue create -R BrainCOGS/RecordingProcessJobGUI`, template `.github/ISSUE_TEMPLATE/agent_bug_report.md`). Put the error exactly as the user pasted it under **Reported error**; don't ask for or invent a Slack link. Then fix it with the `bugfix-from-slack` skill (agent-tooling); the PR says `Fixes #<issue>`. Commits, pushes, issues and PRs all go through the user's account.
 - Commits end with one `Assisted-by: <Harness>:<model>` trailer per agent, e.g. `Assisted-by: ClaudeCode:claude-opus-5-5`. No AI co-authors, and no `Generated with` lines. `prek install` sets up the `ai-attribution` commit-msg hook (`.github/scripts/check_ai_attribution.py`, vendored from agent-tooling), and CI checks every commit of a PR.
 - PRs target `main` on `BrainCOGS/RecordingProcessJobGUI`, opened from the user's own account.