skills.rest
sohamda
sohamdaCOMMUNITY
Explore all skills by this author →
Website𝕏@iamsohamGitHub

vendor-prompting

Audit agent and prompt files against Anthropic Claude and OpenAI GPT-5.5 prompting best practices.

Updated May 12, 2026
Search Tags:#linting#prompt engineering#claude#anthropic#agent audit#gpt-5.5#frontmatter validation
One-click install
npx skills add https://github.com/sohamda/apex-try-out-demo-repo --skill vendor-prompting-sohamda
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: vendor-prompting
Source: https://github.com/sohamda/apex-try-out-demo-repo/tree/main/.github/skills/vendor-prompting
Command: npx skills add https://github.com/sohamda/apex-try-out-demo-repo --skill vendor-prompting-sohamda

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) and assets (resource) components.

What problem does it solve? Authoring and reviewing .agent.md and .prompt.md files for different model vendors requires tracking divergent prompting conventions (Claude XML structuring vs GPT-5.5 outcome-first skeletons), and manual reviews are inconsistent without a canonical, machine-readable rule set. ## Core Features & Use Cases - Vendor rule catalog: A rules.json source of truth maps every rule to an ID, severity, source citation, and validator check, covering Claude XML blocks, GPT-5.5 skeleton sections, handoff enrichment, and frontmatter model style. - Six-step audit procedure: Read frontmatter, classify the model family, load the matching checklist, run validate-agents.mjs, perform a manual pass, and produce a verdict report (APPROVED / NEEDS_REVISION / REJECTED) using the audit template. - Model-family detection: Mirrors the validator's classifyModel() logic to assign families (claude-opus, claude-sonnet, gpt-5.5, etc.) with per-family severity statuses. - Use Case: A reviewer investigating an npm run lint:vendor-prompting finding loads this skill, maps the ruleId to its rule and vendor source, then follows the checklist to confirm or remediate the violation. ## Quick Start Ask the agent to audit a specific .agent.md file against vendor prompting best practices and produce a verdict report.

Frequently Asked Questions about vendor-prompting

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I audit a .agent.md file against vendor prompting best practices?▼

Follow the six-step audit procedure: read the frontmatter, classify the model family, load the matching checklist from checklists.md, run validate-agents.mjs with --only=vendor-prompting, perform a manual pass, and produce a report using the audit template. The verdict is APPROVED if there are zero errors and at most five warnings.

What prompting rules differ between Claude and GPT-5.5 agents?▼

Claude agents use XML structuring blocks like <investigate_before_answering> and <output_contract>, while GPT-5.5 agents require an outcome-first markdown skeleton with # Goal, # Success criteria, # Constraints, # Output, and # Stop rules sections. GPT agents must not contain Claude-specific XML blocks.

How does the validator classify model families in frontmatter?▼

The classifyModel() function lower-cases the model value and matches substrings in priority order, such as 'claude opus' for claude-opus or 'gpt-5.5' for gpt-5.5. For agents with a model array, the first entry decides the family, and an unclassifiable value raises an error.

Why does my agent fail validation with frontmatter-model-style-001?▼

This error means the model field uses the wrong YAML form: .agent.md files must use array form and .prompt.md files must use string form. Bareword values with parenthetical qualifiers are forbidden because YAML misparses them.

When should I not load the vendor-prompting skill?▼

Do not load it for routine prompt edits where the format is already known; the auto-loaded vendor-prompting.instructions.md file carries the hard-rule shortlist for those cases. It is also out of scope for generic markdown style issues.

Related Skills

Explore More Software Engineering Skills→
selfagencyselfagency

model-recommendation

Analyze .agent.md or .prompt.md files to recommend optimal GitHub Copilot models.

Community
Intermediate
codercoder
2.0k

deep-review

Coordinate specialized sub-agents to review code changes and consolidate findings.

Official
Advanced
kdlbskdlbs
562

verify

Run format, typecheck, tests, and lint across monorepo packages.

Official
Intermediate
pazdedavpazdedav

copilot-customization

Guide VS Code Copilot customization mechanisms including instructions, prompt files, agents, and MCP servers.

Community
Intermediate