Bridge Actions
Bridge can take actions across your entire project. This page catalogs every action Bridge can perform, organized by resource type and action category.
Action Categories
| Category | Behavior | Examples |
|---|---|---|
| Read | Executes immediately, no confirmation | List agents, check execution status |
| Write | Shows preview card, requires Confirm click (Safe Mode) or executes immediately (Auto Mode) | Create agent, update workflow |
| Destructive | Always requires explicit confirmation, even in Auto Mode | Delete agent, cancel execution |
Agent Actions
Read
| Prompt | What Bridge Does |
|---|---|
| "List my agents" | Shows all agents in the project with name, model, and capabilities |
| "What does the researcher agent do?" | Shows full agent config: system prompt, model, capabilities, skills, RAGs, MCPs |
| "Which agents use gpt-4o?" | Filters agents by model |
| "Which agents have web_search enabled?" | Filters by capability |
Write
| Prompt | What Bridge Does |
|---|---|
| "Create an agent named summarizer using gpt-4o-mini" | Creates agent with specified name and model |
| "Create a researcher agent with web_search and web_scraper enabled, using claude-sonnet-4-6" | Creates agent with capabilities |
| "Update the writer agent to use gemini-2.0-flash" | Changes model |
| "Enable web_scraper on the analyzer agent" | Enables a capability |
| "Change the researcher's system prompt to: You are a senior analyst..." | Updates system prompt |
| "Add the SEO Writer skill to the content-writer agent" | Assigns skill |
| "Set the writer agent's temperature to 0.3" | Updates temperature |
| "Add the company-docs RAG source to the support-agent" | Assigns RAG source |
Destructive
| Prompt | What Bridge Does |
|---|---|
| "Delete the researcher agent" | Permanently deletes agent (shows confirmation) |
| "Remove all capabilities from the summarizer" | Removes all capability toggles (shows preview) |
Workflow Actions
Read
| Prompt | What Bridge Does |
|---|---|
| "Show me all workflows" | Lists workflows with name, trigger type, last run |
| "Describe the content pipeline workflow" | Shows node structure, connections, trigger config |
| "What workflows use the researcher agent?" | Finds workflows referencing a specific agent |
| "Which workflows haven't run in the last week?" | Filters by last execution date |
Write
| Prompt | What Bridge Does |
|---|---|
| "Create a workflow that uses researcher then writer" | Creates workflow with both agent nodes connected |
| "Add a Branch node after the analyzer in the security workflow" | Adds node to existing workflow |
| "Set the security monitor to run every day at 9am EST" | Changes trigger to schedule |
| "Add a webhook trigger to the content pipeline" | Adds webhook trigger |
| "Connect the researcher to the writer in the research workflow" | Adds edge between nodes |
Destructive
| Prompt | What Bridge Does |
|---|---|
| "Delete the test-workflow" | Permanently deletes workflow |
| "Remove the old analyzer node from the security workflow" | Removes node from workflow |
Execution Actions
Read
| Prompt | What Bridge Does |
|---|---|
| "Show me recent executions" | Lists last 10 executions across all workflows |
| "Show yesterday's failures" | Filters by date and failed status |
| "What did the content pipeline produce last time it ran?" | Shows most recent execution output |
| "How long do the security monitor runs take?" | Shows execution duration stats for a workflow |
| "Are there any pending approvals?" | Lists all paused HiL executions |
Write
| Prompt | What Bridge Does |
|---|---|
| "Run the content pipeline" | Triggers workflow execution using the trigger's default input parameters |
| "Run the content pipeline with topic: AI in education" | Triggers execution, overriding the topic param (other params keep their defaults) |
| "Approve the pending code review" | Approves the most recent HiL approval |
| "Reject the pending deployment with message: Found security issue in line 42" | Rejects HiL with comment |
| "Re-run the last failed content pipeline execution" | Triggers re-run of most recent failed execution |
Trigger parameters & live execution panel
Running a workflow from Bridge behaves exactly like the Run ▶ button: it sends the trigger's default input parameters, which you can override in natural language. In both Safe and Auto confirmation modes the run shows a live execution panel in the chat (and in the Executions page) so you can watch it progress.
Destructive
| Prompt | What Bridge Does |
|---|---|
| "Cancel the running execution" | Cancels current running execution |
RAG Source Actions
Read
| Prompt | What Bridge Does |
|---|---|
| "List my RAG sources" | Shows all RAG configs with provider and connection status |
| "Is the company-docs RAG source connected?" | Shows connection status and document count |
Write
| Prompt | What Bridge Does |
|---|---|
| "Create a Pinecone RAG source named product-docs with API key [key] and index product-index" | Creates RAG config |
| "Update the company-docs RAG to use top_k=5" | Updates retrieval configuration |
Destructive
| Prompt | What Bridge Does |
|---|---|
| "Delete the product-docs RAG source" | Permanently deletes RAG config |
MCP Server Actions
Read
| Prompt | What Bridge Does |
|---|---|
| "Show connected MCP servers" | Lists all MCP servers with tool counts |
| "What tools does the github MCP provide?" | Lists discovered tools from a server |
Write
| Prompt | What Bridge Does |
|---|---|
| "Connect a GitHub MCP server with token [token]" | Adds MCP server connection |
Skill Actions
Read
| Prompt | What Bridge Does |
|---|---|
| "What skills are available?" | Lists all skills (builtin + custom) |
| "Show me the security analyst skill" | Shows skill details and system prompt |
Write
| Prompt | What Bridge Does |
|---|---|
| "Create a skill called brand-voice with system prompt: Always use active voice..." | Creates custom skill |
Settings Actions
Read
| Prompt | What Bridge Does |
|---|---|
| "Who are the members of this project?" | Lists members with roles |
| "What API keys do I have?" | Lists LLM key names and providers |
Write
| Prompt | What Bridge Does |
|---|---|
| "Invite sarah@company.com as an editor" | Sends project invitation |
| "Change Bob's role to reader" | Updates member role |
Information / Documentation Actions
Always Read (no confirmation needed)
| Prompt | What Bridge Does |
|---|---|
| "How do I set up a webhook trigger?" | Explains from TARX docs (RAG retrieval) |
| "What's the expression syntax for loop nodes?" | Provides expression reference |
| "What's the difference between Safe and Auto mode?" | Explains confirmation modes |
| "How does RAG enrichment work?" | Explains the RAG query flow |
Bridge answers documentation questions using TARX's indexed docs. Answers are grounded in actual documentation, not hallucinated.
Multi-Step Action Chains
Bridge handles multi-step requests that involve multiple actions:
"Create an agent called security-analyst with the security-analyst skill and
web_search enabled, then add it to the security-monitor workflow right after
the trigger node, and run the workflow to test it"
Bridge will:
- Create the agent (shows preview → confirm)
- Add node to workflow (shows preview → confirm)
- Run the workflow (shows preview → confirm)
Each step shows its own preview card. You confirm them in sequence.
Action Limits
Bridge enforces safety limits on actions:
| Limit | Value | Reason |
|---|---|---|
| Max agents created per conversation | 10 | Prevents accidental bulk creation |
| Max workflows created per conversation | 5 | Prevents runaway creation |
| Max executions triggered per conversation | 20 | Cost protection |
| Max bulk operations | 1 at a time | Prevents batch destructive actions |
These limits reset when you start a new conversation.
What Bridge Won't Do
Bridge refuses certain operations for safety:
- Deleting all agents or workflows in bulk ("delete all my agents" → asks you to confirm each one)
- Actions that would break other resources (deleting an agent used in active workflows warns you first)
- Accessing or modifying other users' API keys
- Accessing resources from other projects
- Any action that would violate your role permissions (Readers cannot create via Bridge either)