Skip to content

better-isms/isms-copilot

v0.1.8MIT

Account MCP connection that lets a compliance specialist take over GRC work: framework interpretation, policy drafting, control mapping, gap analysis, risk registers and audit prep, with workspaces, documents, memories and company context in one place.

Changelog

0.1.8 (2026-11-04)

  • Review re-submission after the 2026-11-04 rejection ("appears to enable or facilitate commerce for disallowed offerings"). The skill no longer links the pricing page or mentions credits or purchases: a feature that is not on the user's plan is stated plainly, with no links. The two commerce-shaped negative review cases (API key, API credits) are replaced by out-of-scope cases (password change, account deletion). A file-upload case was dropped: run in ChatGPT, the model offered an upload that no tool provides. release notes trimmed to the facts; commerce_description says what the tools carry (no checkout, payment or API key functions, no plan or upgrade links in results).
  • mcp.json declares only auth type oauth; client and endpoints come from discovery, which restores the default OIDC scopes (openid, email, offline_access). Declaring endpoints had reduced the request to email.
  • plugin.json name is isms-copilot: a new OpenAI plugin under the verified business identity, because a plugin's first connection keeps its OAuth client for good.
  • Versions mirrored in the Claude and Cursor manifests. No tool or listing-metadata change; mcp.json gains the auth declaration only; the OpenAI package description is unchanged.

Claude Code 0.1.7 (2026-10-19)

  • README is now the public listing text (the Claude Code plugin directory shows it as the description): the approved Claude connector listing text, what the plugin contains and sends, install for Claude Code, Gemini CLI and other clients, maintainer notes.
  • .claude-plugin/plugin.json: version 0.1.7, description matches the Claude directory listing, homepage carries ref=claude-code-plugins. Its extensions.com.openai block is unchanged. The ChatGPT and Codex package (root plugin.json, mcp.json) stays at 0.1.6.
  • .cursor-plugin/plugin.json (new): Cursor manifest for cursor.directory, description ends "from Cursor.", homepage carries ref=cursor-directory.

Gemini CLI 0.1.0 and Claude Code install (2026-10-14)

  • New files only; the OpenAI package files (plugin.json, .claude-plugin/plugin.json, mcp.json, skills, assets) are unchanged and the package stays at 0.1.6.
  • gemini-extension.json (extension 0.1.0, MCP server ismscopilot by httpUrl) and GEMINI.md context, so the repo installs with gemini extensions install and the Gemini CLI gallery can index it.
  • .mcp.json (type http) and .claude-plugin/marketplace.json, so the repo installs as a Claude Code plugin with claude plugin marketplace add.
  • README: Install section for both.

0.1.6

  • Review: demo_recording_url added (video walkthrough of the review cases in ChatGPT, hosted on www.ismscopilot.com).

0.1.5

  • Review case P4 prompt names both steps ("List my ISMS Copilot workspaces, then list the documents in My First Workspace."). In one of two ChatGPT runs the earlier wording led to a list_workspaces call only and a "no documents" answer. The matching starter prompt now names both steps too ("..., then list the documents in the first one.").

0.1.4

  • Review cases P2 and P3: tools_triggered holds plain tool names (create_conversation, send_message). The conditional get_reply note stays in expected_behavior. Matches the documented format (tool names, comma separated).

0.1.3

ChatGPT test run on the reviewer account (2026-10-07):

  • Review case P3 prompt names the app ("Ask ISMS Copilot to turn that into ..."). The bare follow-up was answered by ChatGPT itself without a tool call; the named prompt continues the same specialist conversation (send_message, verified: one thread, four messages).
  • Negative case N2 expected behavior matches what ChatGPT does: declines without a tool call and says key creation is not available through this connection.

0.1.2

Pre-submission audit against the OpenAI plugin guidelines (2026-10-07):

  • Subtitle "Compliance and GRC specialist" (states the purpose for the Security category; drops "your agent", which could read as a reference to another AI assistant).
  • Long description: restricted-data line (no health records, card data or government IDs), not-legal-advice line, credit-purchase wording removed from the listing (commerce stays declared false).
  • Support URL points to the public contact-support page.
  • composerIcon added (required in the Codex-format manifest the portal derives).
  • Review case P1 now reads the company profile (get_company_context) instead of the plan, so no plan, upgrade link or account identifier appears in review output. P2 and P3 name get_reply as conditional.
  • Onboarding skill: writes only on user request; sensitive data and not-legal-advice guidance.

0.1.1

Category corrected to Security (the ChatGPT plugins directory category list confirmed from the live directory: Security is the category for information security compliance work; Productivity could not be confirmed by the portal check). Version bump to 0.1.1. No other listing text change.

0.1.0

Initial plugin package for the ChatGPT and Codex plugins directory, "With MCP" type. Connects the ISMS Copilot account MCP server (https://account.ismscopilot.com/v1/account/mcp, Streamable HTTP, OAuth 2.1). Listing metadata under extensions.com.openai in both the root Agent Plugins manifest and .claude-plugin/plugin.json: displayName and subtitle within the 30-character limits, no pricing in the description, no references to other AI assistants, four listing URLs, starter prompts, five positive and three negative review cases, and the get started onboarding skill. API key and credit purchases are not part of the connected tool set, and commerce is declared false.