* bootstrapping agent-plugins provider
* nesting into a folder in case we start hosting more than one plugin in the future
* Align Agent Plugin paths with nested layout
Document the multi-plugin provider convention and keep sync automation pointed at the relocated portable package.
Co-authored-by: Cursor <cursoragent@cursor.com>
Committed-By-Agent: cursor
* changing installation instructions
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
* Add Grok Build provider plugin
Ship providers/grok/plugin from the Claude layout with Grok-native wiring:
named Company Researcher subagent (tools allowlist), connect-recommend
spawn path, SKILL_DIR refs, .grok-plugin marketplace, and sync/guard hooks.
* Track Grok provider .mcp.json
.gitignore ignores .mcp.json; force-add like the Claude provider so the
hosted Stripe MCP ships with the plugin.
* Use kebab-case company-researcher agent id
Frontmatter name must match agents/<name>.md so Grok spawn resolve
finds the file. Display name "Company Researcher" broke spawn while
still appearing in Available types.
* company-researcher: Grok-native tools and deny MCP
Use read_file/list_dir/grep/web_fetch/web_search allowlist and
disallowedTools search_tool/use_tool so parent MCP does not leak into
the research subagent.
* Set Grok provider version to 0.1.0
First public publish; start at 0.1.0 like Cursor's early line.
* Simplify Grok Build install to marketplace command
Document grok plugin install stripe --trust for the official
marketplace path once the plugin is listed.
* Shorten Grok plugin description for marketplace
Match the wording requested on the PR review.
* bootstrapping a plugin directory for codex, even if they can't actually use the format in their own marketplace yet
* moving to white-on-blurple parallelogram, matching other icons
* finished setting things up for teh codex directory GIT_VALID_PII_OVERRIDE
* breaking files into newlines for sync-skills, not adding our marketplace file because OpenAI's syntax lets us point to the version file in the relevant directory
After writing skill files, use git status to detect whether skills were
added/deleted (minor bump) or only updated (patch bump), then increment
the version in all four plugin manifests so consumers pick up the changes.
Committed-By-Agent: claude
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* Hardcode connect-recommend as exempt instead of using .local marker mechanism
Replace the generic .local file-based exemption with explicit hardcoded lists
in sync.js (LOCAL_SKILLS set) and guard-skills.yml (path exclusion pattern),
making adding new local skills require an intentional code change.
Committed-By-Agent: claude
* version bump the claude plugin
* Add connect-recommend skill and company-researcher agent to claude plugin
Update sync.js to preserve locally-added plugin directories by skipping
cleanup for any directory containing a .local marker file.
Co-Authored-By: Chris Greer <chrisg@stripe.com>
Co-Authored-By: Emily Littlepage <ezl@stripe.com>
* update the guard workflow to check for a .local mechanism to avoid false positives
---------
Co-authored-by: Chris Greer <chrisg@stripe.com>
Co-authored-by: Emily Littlepage <ezl@stripe.com>
* Updated sync-skills script & workflow to handle new/deleted/renamed files, generally cleared out code that only made sense when skills were manually edited in this repo
* guard-skills: use app token for reviews, exclude skills/README.md
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Committed-By-Agent: claude
* link in my guard comment was actually broken, had to point to index.json
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Restructure stripe-best-practices skill with progressive disclosure
Replaces the single flat SKILL.md with a concise router + domain-specific
reference files. Agents load the short SKILL.md (~200 tokens) at activation
and only read the relevant reference for the task at hand.
Structure:
SKILL.md — routing table + key docs
references/payments.md — Checkout Sessions, PaymentIntents, deprecated APIs, PCI
references/connect.md — Accounts v2, controller properties, charge types
references/billing.md — Billing APIs, Checkout pairing, Customer Portal
references/treasury.md — v2 Financial Accounts
Eval results (6 scenarios, deterministic checks):
No skill: 19/29 (66%)
Flat skill (before): 21/29 (72%)
Progressive disclosure: 24/29 (83%)
Also improves the description with keywords for better skill discovery,
and removes the invalid alwaysApply frontmatter field.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Committed-By-Agent: claude
* Fix content drift: restore Charges API hard prohibition and 'Always describe'
Two pieces of guidance drifted from the original during restructuring:
1. The Charges API prohibition was weakened from "Never recommend the
Charges API" (absolute) to a general deprecated-API escape clause.
Restored as a hard prohibition above the table.
2. "Always describe accounts in terms of..." lost the word "Always".
Restored.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Committed-By-Agent: claude
* Add CI check and sync script for provider skill directories
Adds a CI job that fails PRs if skills/ and the provider plugin
directories (providers/claude/plugin/skills/, providers/cursor/plugin/skills/)
are out of sync. Also adds a helper script to do the copy.
When CI fails, the fix is:
./scripts/sync-skills.sh
Also syncs the progressive disclosure reference files to both provider
directories.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Committed-By-Agent: claude
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
John O'Sullivan