agent-buglisted

# Agent Bug Use this skill when Codex needs to record a concrete bug found during Agent Workbench work, external-agent review, result collection, dry runs, tests, or manual inspection. ## When To Record - A test fails and the failure is not immediately fixed in the same step. - External review reports a concrete defect with evidence. - `Collect-AgentResult.ps1` shows a blocking issue or bug-like signal that Codex confirms is a real defect. - A manual dry run reveals a repeatable failure. ## Rules - Codex remains the primary conversation and final reviewer. - Prefer evidence over interpretation: include command output, file paths, reproduction steps, and expected versus actual behavior. - Keep each bug record focused on one defect. - Do not record vague, duplicate, or secret-containing entries. - Do not paste API keys, provider config, tokens, private launcher contents, or secret-like text into bug fields. - Do not treat result collection warnings as automatic bugs; Codex decides whether to register one. ## Create A Bug Run `New-AgentBug.ps1` from the installed workbench scripts: ```powershell $WorkbenchRoot = $env:AGENT_WORKBENCH_HOME & (Join-Path $WorkbenchRoot "scripts\New-AgentBug.ps1") ` -WorkbenchRoot $WorkbenchRoot ` -Slug "short-bug-slug" ` -Severity medium ` -Source codex ` -Summary "Short concrete bug summary." ` -Evidence "Relevant command output, file reference, or review evidence." ` -Reproduction "Steps or command that triggers the issue." `