这是indexloc提供的服务,不要输入任何密码
Skip to content

Conversation

@K-Mistele
Copy link
Contributor

@K-Mistele K-Mistele commented Oct 23, 2025

What problem(s) was I solving?

ResearchBot (LinearLayer/Claude) was posting extremely verbose comments to Linear tickets containing the entire Claude conversation history, including tool calls, thinking processes, and implementation details. This made Linear comments difficult to read and cluttered the ticket history with unnecessary information.

Issue: ENG-2240 - Update ResearchBot comments to be terse with doc link and open questions

What user-facing changes did I ship?

For Linear ticket viewers:

  • ResearchBot comments are now concise and readable, containing only:
    • "research completed" message with a clickable link to the research document
    • A list of open questions from the research
  • No more verbose conversation logs, tool outputs, or implementation chatter in Linear comments

Example new format:

research completed - [thoughts/shared/2025-10-21-ENG-XXXX-description.md](https://github.com/humanlayer/thoughts/blob/main/repos/humanlayer/shared/research/2025-10-21-ENG-XXXX-description.md)

### open questions
1. Should we add retry logic for API failures?
2. How should we handle legacy data format?
3. None at this time [if no questions]

How I implemented it

Single, Simple Change - Prompt Modification:

Updated the Claude prompt in .github/workflows/linear-research-tickets.yml to explicitly instruct Claude to output ONLY the terse format as its final response.

Key changes:

  1. Removed the old instruction about formatting GitHub links inline
  2. Added a CRITICAL: section that specifies the exact output format required
  3. Emphasized "Output ONLY this message as your final response, nothing else"

What I did NOT do (per the implementation plan):

  • No parsing or extraction logic
  • No markers or separators
  • No modifications to the Linear CLI
  • No changes to /research_codebase command
  • Just a clean prompt change

The workflow still captures output with tee CLAUDE_ANSWER.md and posts it via linear add-comment, but now Claude's output is exactly what we want posted.

How to verify it

Automated Verification:

  • YAML syntax is valid (verified with yq)
  • Workflow file syntax is correct

Manual Testing:

To test this change:

  1. Create or use a Linear ticket with status "research needed" assigned to "LinearLayer (Claude)"
  2. Trigger the workflow manually: gh workflow run "Linear: Research Task" --repo humanlayer/humanlayer
  3. Wait for the workflow to complete
  4. Check the Linear comment - it should contain ONLY:
    • Research completion message with document link
    • Open questions section
    • No verbose conversation history, tool outputs, or extra explanations
  5. Click the research document link to verify it navigates correctly

Note: This change only affects NEW research runs. Existing comments with old format will remain unchanged.

Description for the changelog

ResearchBot now posts concise Linear comments - Updated the automated research workflow to post terse comments containing only the research document link and open questions, eliminating verbose conversation logs from Linear ticket comments.


Additional Changes in This PR

This PR also includes work related to directory validation and creation for the WUI (CodeLayer) and HLD daemon:

  • Added directory validation and creation endpoints to HLD API
  • Added directory validation to WUI launch flow with creation dialog
  • Deferred directory validation for draft sessions to launch time
  • Various cleanup and regeneration of API types

These changes improve the user experience when launching Claude Code sessions with directories that don't exist yet.


Important

Updated ResearchBot to post concise comments in Linear tickets by modifying the Claude prompt in the GitHub workflow.

  • Behavior:
    • ResearchBot comments in Linear are now concise, containing only a research completion message with a document link and open questions.
    • Removed verbose conversation logs and tool outputs from comments.
  • Workflow Changes:
    • Updated .github/workflows/linear-research-tickets.yml to modify the Claude prompt to output only the specified terse format.
    • Added a CRITICAL: section in the prompt to specify the exact output format.
    • Emphasized that only the specified message should be outputted as the final response.
  • Misc:
    • No changes to parsing logic, Linear CLI, or /research_codebase command.

This description was created by Ellipsis for d5dd10c. You can customize this summary. It will automatically update as commits are pushed.

Modify the Claude prompt in the linear-research-tickets workflow to
instruct it to output only a terse format containing:
- Research completion message with document link
- Open questions section

This eliminates verbose conversation history from Linear comments
while preserving essential information.

Implements ENG-2240
Copy link
Contributor

@ellipsis-dev ellipsis-dev bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Important

Looks good to me! 👍

Reviewed everything up to d5dd10c in 1 minute and 1 seconds. Click for details.
  • Reviewed 34 lines of code in 1 files
  • Skipped 0 files when reviewing.
  • Skipped posting 3 draft comments. View those below.
  • Modify your settings and rules to customize what types of comments Ellipsis leaves. And don't forget to react with 👍 or 👎 to teach Ellipsis.
1. .github/workflows/linear-research-tickets.yml:129
  • Draft comment:
    The CLAUDE command string now removes the extra trailing space after ${{ env.TICKET_FILE_PATH }}. Please confirm that enclosing the parameter value in single quotes still allows GitHub’s ${{ }} substitution to work correctly. It’s a minor change, but ensuring variable expansion in this shell context is important.
  • Reason this comment was not posted:
    Comment did not seem useful. Confidence is useful = 0% <= threshold 50% The comment is asking the PR author to confirm that a change in the command string still allows for correct variable expansion. This falls under asking the author to confirm their intention or to ensure behavior, which is against the rules. The comment does not provide a specific suggestion or point out a clear issue with the code.
2. .github/workflows/linear-research-tickets.yml:142
  • Draft comment:
    The updated prompt now includes a CRITICAL section that clearly instructs Claude to output the final message in a terse, exactly formatted way. This is a solid improvement for ensuring consistent output. Verify that this strict format is reliably met across research runs and that no extra text is accidentally appended.
  • Reason this comment was not posted:
    Comment did not seem useful. Confidence is useful = 0% <= threshold 50% This comment is asking the PR author to verify something, which is against the rules. It doesn't provide a specific suggestion or point out a specific issue in the code.
3. .github/workflows/linear-research-tickets.yml:132
  • Draft comment:
    It appears that the string parameter for the command is missing a closing quote. The command starts with a single quote (') after -p, but doesn't have a matching closing quote at the end. Please fix this to avoid potential shell issues.
  • Reason this comment was not posted:
    Comment was on unchanged code.

Workflow ID: wflow_yh0iKl6P8y8lWMFF

You can customize Ellipsis by changing your verbosity settings, reacting with 👍 or 👎, replying to comments, or adding code review rules.

@K-Mistele K-Mistele merged commit 8bd52a2 into main Oct 23, 2025
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants