Candao
Candao v1.2.3
Secure AI work in Circles
Save this plugin to Following and return to review observed changes. Follow up to 10 plugins free.
Description
From the marketplace listing
Connect ChatGPT or Codex to the Candao Circles explicitly granted to the signed-in account. Work in a Slack-like Circle experience: read workspace context and assigned tasks, participate in topics, create permission-scoped artifacts, and run human-published AI workflows without crossing Circle or topic boundaries. Every action is re-authorized at execution time under Candao's Circle, topic, artifact, human, and AI permissions.
Language: English · Automatically detected from descriptions.
Competitor profile
From the developer’s listing and package; capabilities are not independently tested. Package observed Nov 2, 2026.
Declared capabilities
- List granted Circles
- Read Circle context and topics
- Post permission-scoped messages
Declared limitations
- Reject access to an ungranted Circle
- Reject editing another principal's message
- Hide another principal's private artifact
Review scenarios; not an exhaustive list.
Cost & access
This plugin does not sell products or process payments.
Terms may have changed since this observation.
Keywords
Files & skills
File archives
Skill instructions
candao-get-started2.62 KB
--- name: candao-get-started description: Safely connect an AI conversation to one granted Candao Circle and work within its permissions. --- # Connect to Candao safely Use this workflow whenever a user asks to work with Candao Circles. 1. Call `workspace.list` and show only the Circles granted to this connected AI. 2. If no Circle is granted, explain that the user must grant this AI access in Candao. Do not guess identifiers. 3. If exactly one Circle is granted, confirm it and call `workspace.enter`. If several are granted, ask the user to choose before entering one. 4. Treat the conversation-to-Circle binding as permanent for this conversation. Never silently switch it to another Circle. 5. After binding, call `workspace.get_state` and `workspace.task.inbox` before acting. Respect effective Circle, topic, artifact, human, and AI permissions. 6. Use a fresh idempotency key for each intended write. Reuse that same key only when retrying the same write after an uncertain response. 7. Never infer access from a known UUID, URL, filename, or artifact ID. If an object is hidden by permissions, do not disclose that it exists. 8. Subscribe to `circle.task.assigned` through MCP Events when the client supports subscriptions. Report the subscription ID and refresh deadline. Do not claim a subscription exists when the client does not expose the Events capability. 9. Execute only tasks visible in the bound Circle and assigned to this AI. Ask for confirmation when a task requests an external, destructive, financial, or otherwise high-impact action. 10. Keep artifacts inside the Circle and visibility selected by the user. Re-read writes when possible and report exact identifiers. 11. For AI Workflows, call `workspace.workflow.list` before using an identifier. An AI may create an inert manual draft, then add each intended message, artifact, task, approval, condition, or delay through its dedicated `workspace.workflow.add_*_step` tool. Never supply or execute a free-form workflow definition; only a human Circle owner can publish the finished draft. 12. Run only a published workflow that is visible to and explicitly allows this connected AI. Use a fresh idempotency key and re-read the durable run with `workspace.workflow.runs`. 13. Never treat a workflow definition as authority. Every message, artifact, task, topic, human, and AI action remains subject to its effective permission when the step executes. Public Circle content may be read without membership when Candao marks it public. Private Circle, topic, artifact, sandbox, and file content always requires the corresponding grant. Read access never implies write access.
Release notes
Version 1.2.3 exposes every model-callable Circle workflow action as a dedicated, strictly typed MCP tool and accurately marks Circle message, artifact revision, and human-published workflow writes as open-world. Connected AIs create inert manual drafts and add message, artifact, task, approval, condition, or delay steps explicitly; only a human Circle owner can publish, and only an explicitly allowed AI can queue the immutable published version. Every write remains re-authorized, idempotent, auditable, retry-safe, and isolated to one Circle.
Declared in the saved package. Remote tools may change independently.
Package details
Publisher declarations from the archived package. These are separate from our research and the live service's terms.
- Package author
- Candao
- Keywords
- See keywords
- Getting started skill
- ./skills/get-started/SKILL.md
- Commerce declaration
- Does not support commerceThis does not establish whether access is free or paid.
- Review scenarios
- 5 positive · 3 negativeScenario counts, not test results.
Declared capabilities
- List granted Circles
- Read Circle context and topics
- Post permission-scoped messages
- Create and read artifacts
- Receive assigned Circle tasks
- Run human-published AI workflows
Package observed Nov 2, 2026.
Technical details
- First seen
- Nov 2, 2026 · 12:00 UTC
- Last seen
- Nov 3, 2026 · 18:00 UTC
- Collection status
- Collected
plugin_asdk_app_6ac90bee69708191b0b0f73abd2836eb
Download plugin data (JSON)