CtrlK
BlogDocsLog inGet started
Tessl Logo

git-safety

Stops an agent from destroying uncommitted work with git. Requires git status and git diff before any command that discards changes, treats "revert" and "undo" as manual edits, and never runs reset, checkout, clean, stash drop, or force push without explicit approval. Use when a task involves reverting, undoing, resetting, cleaning, rebasing, or force pushing, or before any git command in a repo with uncommitted changes.

78

Quality

98%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

100%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

An exemplary rules-style skill: every line is either an exact command, a boundary condition, or a verification step, with no background explanation and no ambiguity. Validation checkpoints and checklists are baked into both the general workflow and the revert/undo workflow, which is exactly what a destructive-operation skill requires. The only nitpick is the distinctiveness of the broad description trigger, which is a frontmatter rather than body issue.

DimensionReasoningScore

Conciseness

The body is a lean ruleset with zero padding: it never explains what git is or how it works, and even the single motivating sentence ('one `git checkout -- .` erased two days of work') earns its place by justifying the rules' rigidity. Every section is actionable content that assumes Claude's competence.

5 / 5

Actionability

Fully concrete guidance throughout: exact commands (`git status`, `git diff <file>`, `git stash push -m "<why>"`, `git revert <sha>`, `git diff --stat`), a table enumerating each destructive command with precisely what it destroys, a safe-defaults list, and a copy-paste-ready suggested-commit example. Nothing is left as vague direction.

5 / 5

Workflow Clarity

Although this skill governs destructive operations, explicit validation is present everywhere: the pre-command sequence (status → stop if dirty → ask), the diff-before-discard rule with its --stat fallback for long diffs, the revert workflow's read-diff-then-edit steps, and two checklists with verification items ('git status ran and I read it', 'no uncommitted work is in the blast radius, or the user approved'). The destructive-operation cap therefore does not apply, and the content matches the anchor's 'explicit validation steps... checklists' description.

5 / 5

Progressive Disclosure

The skill needs no external references — it is a compact single-purpose ruleset (~83 lines) with clear section headers (before-any-command, approval table, revert-as-edit, diff-before-discard, safe defaults, checklists) that make navigation trivial, and no content that belongs in a separate file. The bundle contains only logo assets, none of which are needed for instruction.

5 / 5

Total

20

/

20

Passed

Description

96%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong description: third-person voice, concrete enumerated behaviors, an explicit 'Use when...' clause with natural trigger terms, and a clearly distinct safety niche. The only weakness is the deliberately broad fallback trigger ('before any git command in a repo with uncommitted changes'), which raises mild conflict risk with other git skills.

Suggestions

Consider narrowing the fallback trigger (e.g., 'before any git command that discards, resets, or rewrites history') to reduce overlap with non-destructive git skills like commit-message generation.

Adding 'rollback' and 'discard my changes' as trigger synonyms would make the natural-phrase coverage airtight.

DimensionReasoningScore

Specificity

The description lists multiple specific concrete actions — 'Requires git status and git diff before any command that discards changes, treats "revert" and "undo" as manual edits, and never runs reset, checkout, clean, stash drop, or force push without explicit approval' — covering both the required pre-checks and the gated commands comprehensively, matching the anchor-5 example's level of coverage.

5 / 5

Completeness

It explicitly answers both what ('Stops an agent from destroying uncommitted work with git. Requires git status and git diff... never runs reset, checkout, clean, stash drop, or force push without explicit approval') and when ('Use when a task involves reverting, undoing, resetting, cleaning, rebasing, or force pushing...'), structurally matching the anchor-5 example exactly.

5 / 5

Trigger Term Quality

The trigger phrase 'Use when a task involves reverting, undoing, resetting, cleaning, rebasing, or force pushing' covers the natural verbs a user would actually say, plus 'before any git command in a repo with uncommitted changes'. Only trivial synonyms (e.g., 'rollback') are absent, which does not pull it below the comprehensive anchor.

5 / 5

Distinctiveness Conflict Risk

The core niche (safety around destructive git operations) is clearly distinct with specific triggers, but the fallback clause 'or before any git command in a repo with uncommitted changes' makes it liable to fire alongside any git-related skill. That is minor overlap risk with closely related skills (anchor 4), not the minimal-conflict clarity of anchor 5.

4 / 5

Total

19

/

20

Passed

Validation

100%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository
waynesutton/convexskills
Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.