Skip to content

因境施法,场景先行 · Redesign SKILL.md with scenario-driven workflows - #30

Merged
eric8810 merged 2 commits into
mainfrom
fix/skill-creator-conformance
Jul 22, 2026
Merged

eric8810 merged 2 commits into
mainfrom
fix/skill-creator-conformance

Conversation

@eric8810

Copy link
Copy Markdown
Contributor

Summary

Redesign the local-ocr Agent Skill to follow skill-creator conventions with scenario-driven workflows.

Changes

SKILL.md

  • Add YAML frontmatter (
    ame + description) as required by skill-creator spec — this is the triggering mechanism
  • Move 'when to use' triggers into description frontmatter (body only loads after triggering)
  • Replace command-listing with real usage scenarios:
    • Screenshot with small text: full recognize → ROI re-run if low confidence
    • Form/receipt field extraction: detect-then-recognize two-step pattern
    • Counting text regions: detect + pipe to parse
    • Verifying multimodal model output: deterministic OCR cross-check
    • Batch processing: shell loop with JSONL
  • Add decision flow diagram for command selection
  • Use imperative form throughout
  • Reduce to ~90 lines (under 500 limit)

agents/openai.yaml (new)

  • display_name, short_description, default_prompt per skill-creator spec

Validation

  • Frontmatter: name + description ✓
  • Name format: lowercase + hyphens ✓
  • Body lines: 90 < 500 ✓
  • No extraneous files (README, CHANGELOG, etc.) ✓

Fixes the SKILL.md from PR #29 which was command-listing without scenario design.

chenghan added 2 commits July 22, 2026 11:18
- Add YAML frontmatter (name + description) as required by skill-creator
- Move 'when to use' triggers into description frontmatter
- Use imperative form throughout
- Reduce body to 88 lines (under 500 limit)
- Add agents/openai.yaml with display_name, short_description, default_prompt
- Remove verbose sections that Codex already knows
…lows

- Replace command-listing with real usage scenarios:
  screenshot small text, form/receipt field extraction (detect-then-recognize
  two-step pattern), counting regions, verifying multimodal output, batch
  processing
- Add decision flow diagram for command selection
- Keep under 500 lines, imperative form, skill-creator spec compliant
@eric8810
eric8810 merged commit 288f225 into main Jul 22, 2026
10 checks passed
birhantprkc pushed a commit to birhantprkc/light-ocr that referenced this pull request Jul 29, 2026
…s#30)

* docs(skill): 遵规重塑,简而可行 · conform SKILL.md to skill-creator spec

- Add YAML frontmatter (name + description) as required by skill-creator
- Move 'when to use' triggers into description frontmatter
- Use imperative form throughout
- Reduce body to 88 lines (under 500 limit)
- Add agents/openai.yaml with display_name, short_description, default_prompt
- Remove verbose sections that Codex already knows

* docs(skill): 因境施法,场景先行 · redesign SKILL.md with scenario-driven workflows

- Replace command-listing with real usage scenarios:
  screenshot small text, form/receipt field extraction (detect-then-recognize
  two-step pattern), counting regions, verifying multimodal output, batch
  processing
- Add decision flow diagram for command selection
- Keep under 500 lines, imperative form, skill-creator spec compliant

---------

Co-authored-by: chenghan <chenghan@fazhitech.cn>
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