feat: improve 5 lowest-scoring skill definitions#50404
feat: improve 5 lowest-scoring skill definitions#50404rohan-tessl wants to merge 1 commit intoDataDog:mainfrom
Conversation
| Skill | Before | After | Change | |------------------------|--------|-------|--------| | omnibus-to-bazel | 16% | 90% | +74% | | allium | 45% | 90% | +45% | | create-core-check | 62% | 90% | +28% | | create-pr | 62% | 89% | +27% | | create-status-provider | 62% | 91% | +29% |
|
I have read the CLA Document and I hereby sign the CLA You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 80a311d92f
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
| metadata: | ||
| disable-model-invocation: true |
There was a problem hiding this comment.
Placing disable-model-invocation under metadata prevents Claude Code from applying it, because this flag is only recognized as a top-level frontmatter field. In this skill, that means /create-pr is no longer protected as a manual-only workflow and can be auto-invoked by the model, which is risky for a command that can create and submit real pull requests.
Useful? React with 👍 / 👎.
| metadata: | ||
| allowed-tools: "Bash, Read, Write, Edit, Glob, Grep, AskUserQuestion" | ||
| argument-hint: "[check-name]" |
There was a problem hiding this comment.
allowed-tools and argument-hint were moved into a nested metadata map, but Claude Code only reads them as top-level frontmatter keys. As a result this skill loses tool pre-approval and slash-command argument hints, which degrades the guided workflow and can block execution with extra permission prompts; the same nesting pattern appears in other edited skills in this commit.
Useful? React with 👍 / 👎.
1 participant
Hey @aiuto 👋
I ran your skills through
tessl skill reviewat work and found targeted improvements in your skills. Here's the before/after:These were easy changes to bring the skill's structure and activation in line with what performs well against Anthropic's best practices.
What does this PR do?
Improves the 5 lowest-scoring
.claude/skills/definitions in the repo.What changed in omnibus-to-bazel
Rewrote description as a quoted string. The original contained
<name>which was parsed as XML tags, breaking validation (caused the 16% score). Added "Use when" clause with trigger terms (migrate, convert, add dependency, omnibus to Bazel). Movedargument-hintandallowed-toolsintometadatablock. All step-by-step content preserved, the skill body was already excellent.What changed in allium
Replaced marketing tagline description with a functional description including "Use when" clause. Moved
versionandauto_triggerintometadata. Collapsed detailed syntax reference sections to brief summaries with links toreferences/language-reference.md. Added explicit authoring workflow section. Reduced from ~220 to ~70 lines while preserving all domain expertise via progressive disclosure.What changed in create-core-check
Added "Use when" clause with trigger terms. Moved
allowed-toolsandargument-hintintometadata. Removed redundant Sender Methods Reference table and Important Notes section (key rules already covered in step-by-step instructions).What changed in create-pr
Added "Use when" clause with trigger terms ("open a PR", "submit code for review"). Moved
disable-model-invocation,allowed-tools, andargument-hintintometadata. Removed redundant PR Description Guidelines section that duplicated step 10.What changed in create-status-provider
Added "Use when" clause with trigger terms. Moved
allowed-toolsandargument-hintintometadata. Added minimal Go code skeleton for provider implementation.Motivation
Skill quality directly affects how reliably agents select and execute them. Low-scoring skills are often skipped by the agent router or produce weaker outputs because the description doesn't match what users actually say, or the body explains concepts the model already knows instead of giving it concrete steps.
In addition, I stress-tested your
omnibus-to-bazelskill against a few real-world scenarios, and it held up really well. This means that your skill meaningfully improves agent steering and contributes to stronger output quality. Kudos for that!Describe how you validated your changes
Ran
tessl skill reviewon all 5 skills before and after (scores above). Verified no content was lost, all step-by-step procedures, code templates, and domain expertise preserved. Changes are SKILL.md documentation only, no code, config, or behavior changes.Additional Notes
Honest disclosure, I work at @tesslio where we build tooling around skills like these. Not a pitch, just saw room for improvement and wanted to contribute.
If you want to self-improve your skills, or define your own scenarios to pressure test, just ask your agent (Claude Code, Codex, etc.) to evaluate and optimize your skill with Tessl. Ping me @rohan-tessl, if you hit any snags.