chore(skills): Improve triage-issue skill (fix false-positives, check changelog, model)#21257
Conversation
| Do NOT use `python3 -c` or other inline Python in Bash, only the provided scripts are allowed. | ||
| Do NOT attempt to delete (`rm`) temporary files you create. | ||
| claude_args: | | ||
| --model claude-opus-4-6 |
There was a problem hiding this comment.
Doesn't this need even more tokens? 😅
There was a problem hiding this comment.
should be way more efficient on token usage than 4.7
| Do NOT attempt to delete (`rm`) temporary files you create. | ||
| claude_args: | | ||
| --model claude-opus-4-8 | ||
| --max-turns 50 --allowedTools "Write,Bash(gh api *),Bash(gh pr list *),Bash(npm info *),Bash(npm ls *),Bash(python3 .claude/skills/triage-issue/scripts/post_linear_comment.py *),Bash(python3 .claude/skills/triage-issue/scripts/parse_gh_issues.py *),Bash(python3 .claude/skills/triage-issue/scripts/detect_prompt_injection.py *),Bash(python3 .claude/skills/triage-issue/scripts/write_job_summary.py *)" |
There was a problem hiding this comment.
Bug: The triage-issue workflow is missing Bash(grep *) and Bash(head *) in its allowedTools list, but the SKILL.md instructs the agent to use these commands for changelog investigation.
Severity: HIGH
Suggested Fix
Add Bash(grep *) and Bash(head *) to the allowedTools list in .github/workflows/triage-issue.yml to permit the agent to execute the commands required for the changelog check.
Prompt for AI Agent
Review the code at the location below. A potential bug has been identified by an AI
agent. Verify if this is a real issue. If it is, propose a fix; if not, explain why it's
not valid.
Location: .github/workflows/triage-issue.yml#L78
Potential issue: The new changelog investigation step, defined in
`.agents/skills/triage-issue/SKILL.md`, instructs the agent to execute a command using
`grep` and `head`. However, the corresponding workflow file,
`.github/workflows/triage-issue.yml`, does not include `Bash(grep *)` or `Bash(head *)`
in its `allowedTools` configuration. When the agent attempts to run the specified
command, the action will block it because the tools are not whitelisted. This will cause
the changelog investigation feature to fail every time it is triggered, rendering this
new functionality inoperative.
Also affects:
.agents/skills/triage-issue/SKILL.md:87~89
Did we get this right? 👍 / 👎 to inform future reviews.
| Do NOT attempt to delete (`rm`) temporary files you create. | ||
| claude_args: | | ||
| --model claude-opus-4-8 | ||
| --max-turns 50 --allowedTools "Write,Bash(gh api *),Bash(gh pr list *),Bash(npm info *),Bash(npm ls *),Bash(python3 .claude/skills/triage-issue/scripts/post_linear_comment.py *),Bash(python3 .claude/skills/triage-issue/scripts/parse_gh_issues.py *),Bash(python3 .claude/skills/triage-issue/scripts/detect_prompt_injection.py *),Bash(python3 .claude/skills/triage-issue/scripts/write_job_summary.py *)" |
There was a problem hiding this comment.
Bug: The triage-issue.yml workflow is missing Grep and Read in its allowedTools list, which are required by the new changelog investigation feature and will cause it to fail.
Severity: CRITICAL
Suggested Fix
Add Grep(./**) and Read(./**) to the allowedTools list in .github/workflows/triage-issue.yml to permit the agent to use the tools required for the new changelog investigation feature.
Prompt for AI Agent
Review the code at the location below. A potential bug has been identified by an AI
agent. Verify if this is a real issue. If it is, propose a fix; if not, explain why it's
not valid.
Location: .github/workflows/triage-issue.yml#L78
Potential issue: The pull request introduces a new "Changelog investigation" feature in
`SKILL.md` that requires the agent to use the `Grep` and `Read` tools. However, the
`allowedTools` list in the `triage-issue.yml` workflow was not updated to include
`Grep(./**)` and `Read(./**)`. The instructions explicitly forbid using `Bash grep/head`
as an alternative. When an issue mentioning a version number is processed, the agent
will attempt to use these un-whitelisted tools, causing the triage workflow to fail.
Also affects:
.agents/skills/triage-issue/SKILL.md:88~90
There was a problem hiding this comment.
The skill can use the Read tool and does not need grep but this was now also updated in the skills markdown to explicitly state this.
Some fixes because this issue was detected as "prompt injection": #21256
Fixes false-positives in the prompt injection lookup.
The System credentials path pattern matched
/root/anywhere it appeared in a string, which caused false positives on Bun's virtual filesystem paths (e.g./$bunfs/root/mainin stack traces) and any embedded path segment likesome/root/folder.Adds a negative lookbehind
(?<!\w)so the pattern only triggers when/root/appears at the start of a path.Add changelog check
Adds the changelog investigation explicitly to the skill as this contains valuable information.
Model Change
Previously, the default model Claude Opus 4.7 was used. This is now changed to Opus 4.8.