docs: clarify Antigravity hook support#126
Conversation
Signed-off-by: clancy <imclancypro@gmail.com>
|
/rosetta Review and triage |
Rosetta Triage ReviewSummary: This PR clarifies that Antigravity and OpenCode integrate with Rosetta via MCP only, and that native hook support is IDE-specific and unverified for Antigravity — addressing an inconsistency between Findings:
Suggestions (non-blocking):
Labels added: Automated triage by Rosetta agent |
|
I have a follow-up patch for this PR but can't push to the fork. This tightens the MCP vs native-hooks wording and updates the mirrored web docs so the public site stays aligned. Why:
Patch: diff --git a/OVERVIEW.md b/OVERVIEW.md
index 4496bdb3..a4dcbb97 100644
--- a/OVERVIEW.md
+++ b/OVERVIEW.md
@@ -17,7 +17,7 @@ Rosetta is a **meta-prompting, context engineering, and centralized instructions
Design principles:
-**Agent-agnostic.** Works across Cursor, Claude Code, VS Code, Windsurf, JetBrains (Copilot, Junie), GitHub Copilot, Codex, Antigravity, OpenCode, and any MCP-compatible IDE. Adopts agent-specific features where available, simulates them where not.
+**Agent-agnostic.** Works through MCP across Cursor, Claude Code, VS Code, Windsurf, JetBrains (Copilot, Junie), GitHub Copilot, Codex, Antigravity, OpenCode, and any MCP-compatible IDE. Adopts agent-specific features where available; native hooks remain IDE-specific and require per-IDE validation.
**Progressive disclosure.** Instructions load in stages (bootstrap, classification, workflow-specific, sub-instructions) to [prevent context overflow](docs/ARCHITECTURE.md#context-overflow-prevention). The agent gets only what it needs for the current task.
diff --git a/docs/CONTEXT.md b/docs/CONTEXT.md
index 62c79a58..43e9aa08 100644
--- a/docs/CONTEXT.md
+++ b/docs/CONTEXT.md
@@ -104,7 +104,7 @@ Not reactive like gateways. Not static like prompt libraries. Verified, project-
## Domain and Operating Context
- Rosetta operates across the full Software Development Lifecycle: planning, requirements, implementation, QA, release, and operations
-- Agent-agnostic: works through MCP with Cursor, Claude Code, VS Code, Windsurf, JetBrains (Copilot, Junie), GitHub Copilot, Codex, Antigravity, OpenCode, and any MCP-compatible IDE. Native hook support is IDE-specific; Antigravity hooks remain unverified and should not be treated as supported until validated.
+- Agent-agnostic: works through MCP with Cursor, Claude Code, VS Code, Windsurf, JetBrains (Copilot, Junie), GitHub Copilot, Codex, Antigravity, OpenCode, and any MCP-compatible IDE. Native hook support is IDE-specific; Antigravity and OpenCode hooks are not currently supported or tested.
- Integrates via [Model Context Protocol (MCP)](ARCHITECTURE.md#rosetta-mcp), the standard transport for AI agent instructions
- Security model: Rosetta is designed to only serve knowledge and instructions to agents — it does not see or process source code. No code leaves the organization's perimeter by design. Rosetta itself does not store any project data. Opt-in features (project datasets, plan_manager, submit_feedback, usage analytics) can be enabled in your deployment for cross-project intelligence, execution tracking, and adoption visibility; when enabled, your infrastructure stores the data and you are responsible for it. plan_manager receives execution plans created by AI, which may contain project-specific information. Usage analytics (PostHog) collects basic operational metadata — IP address, user email, coding agent with version, tool called, and tool parameters — when you deploy and configure a PostHog instance on your infrastructure.
- Grounded in production experience, not theory. Active feedback loop from real projects.
diff --git a/docs/web/docs/architecture.md b/docs/web/docs/architecture.md
index 253af7e5..c739abd3 100644
--- a/docs/web/docs/architecture.md
+++ b/docs/web/docs/architecture.md
@@ -33,6 +33,7 @@ The instructions repo defines *how agents should behave*. The target repo is *wh
│ Target Repository + IDE │
│ Cursor · Claude Code · VS Code · JetBrains · Codex │
│ Windsurf · Antigravity · OpenCode │
+│ (MCP integration; native hooks vary by IDE) │
│ │ │
│ MCP Protocol │
│ (Streamable HTTP + OAuth) │
@@ -167,7 +168,7 @@ Command aliases are used exclusively for Rosetta MCP resources (instructions, kn
Instructions never call MCP tools directly. Rosetta defines command aliases that work across all IDEs and coding agents. This serves three purposes:
-- **Portability.** Same instructions work in Cursor, Claude Code, VS Code, JetBrains, Codex, and any MCP-compatible tool.
+- **Portability.** Same instructions work in Cursor, Claude Code, VS Code, JetBrains, Codex, and any MCP-compatible tool. Native hook support is IDE-specific and must be validated separately.
- **Decoupling.** Instruction content is independent of MCP API changes.
- **Authoring.** Workflows, skills, and rules reference each other through aliases, not tool calls.
diff --git a/docs/web/docs/context.md b/docs/web/docs/context.md
index eaacdd07..9b72f9b4 100644
--- a/docs/web/docs/context.md
+++ b/docs/web/docs/context.md
@@ -112,7 +112,7 @@ Not reactive like gateways. Not static like prompt libraries. Verified, project-
## Domain and Operating Context
- Rosetta operates across the full Software Development Lifecycle: planning, requirements, implementation, QA, release, and operations
-- Agent-agnostic: works with Cursor, Claude Code, VS Code, Windsurf, JetBrains (Copilot, Junie), GitHub Copilot, Codex, Antigravity, OpenCode, and any MCP-compatible IDE
+- Agent-agnostic: works through MCP with Cursor, Claude Code, VS Code, Windsurf, JetBrains (Copilot, Junie), GitHub Copilot, Codex, Antigravity, OpenCode, and any MCP-compatible IDE. Native hook support is IDE-specific; Antigravity and OpenCode hooks are not currently supported or tested.
- Integrates via [Model Context Protocol (MCP)](/rosetta/docs/architecture/#rosetta-mcp), the standard transport for AI agent instructions
- Security model: Rosetta is designed to only serve knowledge and instructions to agents — it does not see or process source code. No code leaves the organization's perimeter by design. Rosetta itself does not store any project data. Opt-in features (project datasets, plan_manager, usage analytics) can be enabled in your deployment for cross-project intelligence, execution tracking, and adoption visibility; when enabled, your infrastructure stores the data and you are responsible for it. plan_manager receives execution plans created by AI, which may contain project-specific information. Usage analytics (PostHog) collects basic operational metadata — IP address, user email, coding agent with version, tool called, and tool parameters — when you deploy and configure a PostHog instance on your infrastructure.
- Grounded in production experience, not theory. Active feedback loop from real projects.
diff --git a/docs/web/docs/overview.md b/docs/web/docs/overview.md
index d316844d..93e877f4 100644
--- a/docs/web/docs/overview.md
+++ b/docs/web/docs/overview.md
@@ -25,7 +25,7 @@ Rosetta is an **open-source governance and context layer** for AI coding agents.
Design principles:
-**Agent-agnostic.** Works across Cursor, Claude Code, VS Code, Windsurf, JetBrains (Copilot, Junie), GitHub Copilot, Codex, Antigravity, OpenCode, and any MCP-compatible IDE. Adopts agent-specific features where available, simulates them where not.
+**Agent-agnostic.** Works through MCP across Cursor, Claude Code, VS Code, Windsurf, JetBrains (Copilot, Junie), GitHub Copilot, Codex, Antigravity, OpenCode, and any MCP-compatible IDE. Adopts agent-specific features where available; native hooks remain IDE-specific and require per-IDE validation.
**Progressive disclosure.** Instructions load in stages (bootstrap, classification, workflow-specific, sub-instructions) to [prevent context overflow](/rosetta/docs/architecture/#context-overflow-prevention). The agent gets only what it needs for the current task. |
isolomatov-gd
left a comment
There was a problem hiding this comment.
See above for diff of changes required to be applied
Rosetta Triage Review — Follow-up Patch AnalysisThanks @isolomatov-gd. The proposed patch is valid and addresses a real gap in PR #126. Here's the analysis: What PR #126 already covers:
What the follow-up patch adds (all valid):
Concerns: None. All changes are factually accurate — the hook verification work (cross-IDE output-format verification series) confirmed that Antigravity and OpenCode hooks are not validated. The wording "not currently supported or tested" is more direct and less ambiguous. Suggestion for the PR author: Please apply this patch to your branch. The web mirror updates are the most important missing piece — without them the public site (docs/web) will diverge from the source docs after merge. Automated triage by Rosetta agent |
|
Thanks for the follow-up patch. I applied the requested docs and web mirror updates in commit Local checks after the update:
Please re-review when you have time. |
Summary
docs/CONTEXT.mdso Antigravity hooks are not implied to be supported whilesrc/hooks/tests/MANUAL-TEST-CASES.mdstill says they are not tested/supported.Fixes #94.
Validation
git diff --checkrg -n "Antigravity|native hooks|Native hook support|hooks are not supported" docs/ARCHITECTURE.md docs/CONTEXT.md src/hooks/tests/MANUAL-TEST-CASES.md