Skip to content

refactor: improve skill descriptions using proven patterns#55

Closed
ozsay wants to merge 2 commits intomainfrom
work-on-triggers
Closed

refactor: improve skill descriptions using proven patterns#55
ozsay wants to merge 2 commits intomainfrom
work-on-triggers

Conversation

@ozsay
Copy link
Collaborator

@ozsay ozsay commented Feb 1, 2026

Summary

Rewrote base44-cli and base44-sdk descriptions following patterns from top skills on skills.sh:

  • vercel-react-best-practices (78K installs)
  • web-design-guidelines (59K installs)

Changes

  • Use "Use when" + verb-based triggers (initializing, configuring, deploying)
  • Use "Triggers on tasks involving" + concrete items
  • Add clear handoff between skills for configuration vs SDK usage
  • Remove aggressive language (AUTHORITATIVE, NEVER assume)
  • Simplify structure from numbered lists to flowing prose

Problem Addressed

Claude didn't activate the CLI skill when implementing agent features, leading to incorrect guidance about dashboard configuration instead of CLI-based agent setup (npx base44 agents push).

Before/After

base44-cli before:

"Use for Base44 CLI operations and project initialization. Triggers: user wants to create/initialize..."

base44-cli after:

"Base44 CLI for project setup and resource configuration. Use when initializing projects, configuring entities/functions/agents, or deploying to Base44. Triggers on tasks involving 'npx base44' commands..."

Test plan

  • Test skill activation with agent-related prompts
  • Verify CLI skill triggers before SDK skill for configuration tasks

Made with Cursor

Rewrote base44-cli and base44-sdk descriptions following patterns from
top skills on skills.sh (vercel-react-best-practices, supabase-postgres).

Changes:
- Use "Use when" + verb-based triggers (initializing, configuring, deploying)
- Use "Triggers on tasks involving" + concrete items
- Add clear handoff between skills for configuration vs SDK usage
- Remove aggressive language (AUTHORITATIVE, NEVER assume)
- Simplify structure from numbered lists to flowing prose

This addresses an issue where Claude didn't activate the CLI skill when
implementing agent features, leading to incorrect guidance about dashboard
configuration instead of CLI-based agent setup.

Co-authored-by: Cursor <cursoragent@cursor.com>
@claude
Copy link
Contributor

claude bot commented Feb 1, 2026

Code review

No issues found. Checked for bugs and CLAUDE.md compliance.

Claude was seeing the skill trigger but choosing not to activate because
it assumed it could infer Base44 config formats from other files. This
adds explicit "MANDATORY: Do not infer" language to both skills.

Blends top-skill patterns (Use when, Triggers on) with strong mandatory
language to address overconfidence in pattern-matching.

Co-authored-by: Cursor <cursoragent@cursor.com>
@ozsay ozsay closed this Feb 1, 2026
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.

1 participant