feat(skills): OpenAI-ready skill set. Workflows as skills, failure modes - #7
eros902002 wants to merge 3 commits into
Conversation
eros902002
commented
Aug 24, 2026
eros902002
commented
- ChatGPT plugins have no slash commands, so the 11 command workflows fold into 7 trigger-described skills (grouped by distinct trigger spaces per OpenAI's one-focused-skill guidance); commands stay for Claude Code
- provider-neutral language throughout; getting-started covers Claude Code, desktop/Cowork, and ChatGPT auth paths
- hypd-tools gains references/failure-modes.md — the public-safe request rules for the classes the HYP-930 baseline measured ChatGPT failing (GAQL change_event, GA4 realtime, research-tool names)
…odes (HYP-930) [AI] - ChatGPT plugins have no slash commands, so the 11 command workflows fold into 7 trigger-described skills (grouped by distinct trigger spaces per OpenAI's one-focused-skill guidance); commands stay for Claude Code - provider-neutral language throughout; getting-started covers Claude Code, desktop/Cowork, and ChatGPT auth paths - hypd-tools gains references/failure-modes.md — the public-safe request rules for the classes the HYP-930 baseline measured ChatGPT failing (GAQL change_event, GA4 realtime, research-tool names) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- microsoft_ads_* and linkedin_* are betaToolAccess: hidden from regular installs, so skills naming them overpromise and mismatch what a review session sees (the HYP-905 trap) - they return with a version bump when the platforms GA Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
There was a problem hiding this comment.
Pull request overview
Updates the HYPD skill set to be more “assistant-agnostic” (suitable for ChatGPT-style skill selection) by replacing slash-command framing with workflow-focused skills, and adds a consolidated reference doc for common validation failures.
Changes:
- Adds new workflow-oriented skills (Google Ads audit, Meta audit, creative analysis, keyword research, reporting/attribution, Shopping feed & landing pages, prompt library).
- Adds
skills/hypd-tools/references/failure-modes.mdand points the core tools skill to it (for faster recovery from validation errors). - Updates onboarding copy (including ChatGPT auth path) and bumps Claude plugin marketplace/manifests to
1.3.0.
Reviewed changes
Copilot reviewed 13 out of 13 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
| skills/hypd-tools/SKILL.md | Updates loading guidance and adds pointer to the new failure-modes reference. |
| skills/hypd-tools/references/failure-modes.md | New reference documenting common validation failures and request constraints. |
| skills/hypd-shopping-feed-and-pages/SKILL.md | New workflow skill for Merchant Center feed audits and landing-page audits. |
| skills/hypd-reporting-attribution/SKILL.md | New workflow skill for monthly reports and GA4 attribution comparisons. |
| skills/hypd-prompt-library/SKILL.md | New workflow skill for browsing and running prompt templates. |
| skills/hypd-overview/SKILL.md | Reframes capabilities around workflows/skills (instead of slash commands). |
| skills/hypd-meta-ads-audit/SKILL.md | New workflow skill for Meta Ads account audits. |
| skills/hypd-keyword-research/SKILL.md | New workflow skill for keyword research / expansion. |
| skills/hypd-google-ads-audit/SKILL.md | New workflow skill covering audit, wasted spend, and negatives. |
| skills/hypd-getting-started/SKILL.md | Updates connection/auth steps to include ChatGPT and provider-neutral wording. |
| skills/hypd-creative-analysis/SKILL.md | New workflow skill for creative analysis (own, draft, or competitor). |
| .claude-plugin/plugin.json | Bumps plugin version to 1.3.0 (manifest metadata). |
| .claude-plugin/marketplace.json | Bumps marketplace plugin version to 1.3.0. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
| "name": "hypd-ai-ads", | ||
| "displayName": "HYPD AI - Paid Ads & Analytics", | ||
| "version": "1.2.5", | ||
| "version": "1.3.0", |
| "name": "hypd-ai-ads", | ||
| "displayName": "HYPD AI - Paid Ads & Analytics", | ||
| "version": "1.2.5", | ||
| "version": "1.3.0", |
| { | ||
| "name": "hypd-ai-ads", | ||
| "version": "1.2.5", | ||
| "version": "1.3.0", |
…or guide (PR #7 review) [AI] - CI requires the version equal across plugin.json, marketplace.json AND server.json — local `claude plugin validate` only checks the first two, which is how 1.2.5 slipped through - server.json describes the remote MCP server for ANY client, so the general connector guide fits it; the plugin.json/marketplace.json homepages stay on the Claude-plugin guide deliberately — they are Claude-facing listing surfaces and that page carries the install commands Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
Thanks Eros, good direction. A few things before merge: 1. Docs URL (blocking): https://docs.hypd.ai/guides/mcp-connector is a 404 right now. Docs live in a separate repo (HYPD-AI/docs, Mintlify), so changing the URL here doesn't create the page. The live walkthrough is https://docs.hypd.ai/guides/claude-code-plugin (200). Did you mean for a new provider-neutral page to exist? We could create that path in the docs repo (duplicate and adapt), or set up a Mintlify redirect. Either way the docs change has to land before this URL ships, or we keep the current URL for now. Which did you have in mind? 2. GAQL nit: failure-modes.md says "no OR across fields", which reads like OR within a field is fine. GAQL WHERE supports only AND, no OR at all. Suggest: "WHERE supports only AND, no OR, no parentheses". 3. Context and a question on the ChatGPT premise: this repo was built for the Claude plugin directory, so it follows the Claude Code plugin format (https://code.claude.com/docs/en/plugins) and is distributed today via the Claude marketplaces and the MCP registry. For cross-assistant, the relevant thing seems to be the Agent Plugins spec (https://github.com/agentplugins/agent-plugins-spec, OpenAI is on the TSC): plugin.json + skills/ + MCP config are the portable interop floor, and slash commands are explicitly left client-specific. So the shape of this PR fits that model well: skills as the portable layer, commands staying Claude-only. Question: do you know how ChatGPT actually consumes these plugins/skills today? The AAIF post (https://aaif.io/blog/from-skills-and-tools-to-portable-agent-plugins) says plugins are complementary with skills and MCP but doesn't specify the consumption path. If there isn't one yet, the PR is still a win for Claude desktop and Cowork (no slash commands there either), I just want to be clear on what we're targeting. 4. Drift between commands and skills: the template-key routing now lives in both commands/ and skills/ (before this PR only commands/ had it). Can we make the commands thin pointers to their skill so keys only ever change in one place? Fine as a follow-up PR. Once merged, 1.3.0 needs the release + registry republish, I'll handle that. |
|
Moving the ChatGPT plugin to https://github.com/HYPD-AI/ads-chatgpt-app |
