fix(skills): instruct agent to use exact absolute path for skill files#1104
Open
buidackim wants to merge 1 commit into
Open
fix(skills): instruct agent to use exact absolute path for skill files#1104buidackim wants to merge 1 commit into
buidackim wants to merge 1 commit into
Conversation
The system prompt had conflicting guidance: the Skills section told the agent to read SKILL.md at the provided location path, while the Workspace section said 'do not guess absolute paths'. Weaker models (e.g. MiniMax-M2.7) would try to reconcile both instructions by stripping the absolute prefix and converting to a relative path, causing 'file not found' errors. This change: - Adds '(use the EXACT absolute path provided)' to skill reading instructions in both inline and search modes - Clarifies workspace guidance: relative paths for user files, exact absolute paths for skills - Updates skill_search result instruction with the same emphasis
mrgoonie
requested changes
Jun 22, 2026
mrgoonie
left a comment
Contributor
There was a problem hiding this comment.
Summary: This is the right problem to fix, but the patch currently breaks the prompt-size test budget in internal/agent.
Risk level: Low
Mandatory gates:
- Duplicate/prior implementation: clear; I only found this PR for the exact skill absolute-path prompt guidance.
- Project standards: issue found — prompt wording changes need to stay within the tested mode size budgets.
- Strategic necessity: clear value; the workspace-vs-skill path ambiguity is real and worth removing.
- CI/checks: missing on GitHub; local targeted test failed.
Findings:
- Important:
go test ./internal/agent ./internal/toolsfails on this branch withTestNoneModeSections:none mode too large: 3112 chars. The added wording is small, but it still pushes an existing prompt-size guard over its limit. Please either tighten the new wording enough to keep the none-mode budget under the test threshold, or update the test budget only if that larger prompt is an intentional product decision.
Verdict: REQUEST_CHANGES
Next step: make the wording more compact (for example, avoid repeating the full “use the EXACT absolute path provided” phrase in multiple emitted sections) and rerun go test ./internal/agent ./internal/tools before re-review.
Posted by /github-maintain at 2026-06-22T21:09:30Z
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
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.
Problem
The system prompt contains conflicting guidance that causes weaker models to mangle skill file paths:
Models like MiniMax-M2.7 try to reconcile both instructions by stripping the absolute prefix from the skill path (e.g.
/app/data/skills-store/gpt-image-2-pro-max/1/SKILL.md→skills/gpt-image-2-pro-max-main/SKILL.md), causingfile not founderrors.Fix
Resolves the contradiction by clarifying the boundary:
(use the EXACT absolute path provided)Files Changed
internal/agent/systemprompt.go— 3 string changes in prompt guidanceinternal/tools/skill_search.go— 1 string change in search result instructionImpact
Minimal prompt size increase (~30 tokens). Helps weaker models follow skill paths correctly without affecting stronger models.