Adding an agent integration

Agent support is built from the agent vendor’s own documentation, never from guesses. docs/agents.md is the source of truth. Each row records:

• the MCP config path and format,
• the hook events and how they block or add context,
• the instructions file,
• the distribution mechanism (plugin, extension, or directory listing),
• a link to the vendor documentation that confirms the row.

The adapter

Each agent is one entry in the registry in src/agents/mod.rs:

• how to detect it: project files, executables on PATH, directories in the home directory;
• where its MCP config lives (project and user-wide) and in which format (McpFormat);
• if it has a hook adapter, its hook file, layout (HookLayout) and events;
• its instruction file, the trust step only the user can take, and what Tessera cannot do for it yet.

tessera init reads the registry; nothing else needs to change for an agent whose config formats already exist.

Steps

1. Confirm the agent’s MCP config location and format, and its hook contract (event names, input JSON, how a hook blocks or adds context), in its documentation.
2. Add or update the agent’s row in docs/agents.md, with the source link. Mark anything the documentation does not confirm as unverified, and do not implement it.
3. Add the registry entry. If its hook output has a new shape, add an Agent variant in src/hook/mod.rs and its arms in render (and silent if it needs JSON on every run). New payload field names go in src/hook/payload.rs.
4. Add tests: a recorded payload from the vendor docs through tessera hook <agent> <event> in tests/hook.rs, and the agent in every_hook_adapter_is_wired_in_its_documented_shape in tests/init.rs.
5. Add the agent to the table in hooks.md and to integrations.md.

Use the agent integration issue template to claim an agent.