Agent Workflows

ShipFoundry turns a relevant stack update into the right next-work artifact, then lets you move that artifact into a user-owned agent workflow without granting automatic implementation authority.

When To Use This

Use this path after ShipFoundry has matched an update to a project and created a briefable output. It is designed for Codex, Claude, ChatGPT, Cursor, and other coding agents that can inspect and verify your local repository.

Do not hand an agent an update title alone. Start with the brief, its evidence, its unresolved questions, and its acceptance checks.

How It Works

ShipFoundry first decides which output lane the available evidence can support. You then choose a delivery path: copy the packet into a local agent, let an authorized MCP client retrieve it, or explicitly export the generated brief to Linear. The user-owned agent verifies the current repository and returns an outcome; ShipFoundry does not run an automatic implementation loop.

Choose the Right Brief

Hand To Agent

• Use When: Existing implementation is proven and the change is bounded.
• Expected Result: A scoped implementation with tests and verification.

Engineering Spike

• Use When: The opportunity is real, but adoption or implementation proof is incomplete.
• Expected Result: A repo-backed finding and a decision on whether to implement.

Research Brief

• Use When: The key unknown needs external or technical research.
• Expected Result: Evidence, remaining unknowns, and a clear stop condition.

Workflow Brief

• Use When: The update changes how the team or system should operate.
• Expected Result: A workflow recommendation with options and triggers.

Strategy Brief

• Use When: The update changes product or commercial direction.
• Expected Result: A decision memo with tradeoffs and a recommendation.

ShipFoundry may show no packet when an update is not useful or grounded enough for a responsible handoff. That is a quality boundary, not an error.

Pick a Delivery Path

Copy to an Agent

Use copy/paste when you want a one-time handoff and the agent already has the right local repository open.

1. Open the generated brief.
2. Check its trust and readiness state.
3. Copy the agent packet.
4. Tell the agent which repository and branch are in scope.
5. Require it to inspect current code before changing anything.
6. Run the acceptance checks and record the outcome in ShipFoundry.

Pull Through MCP

Use MCP for recurring workflows when the agent should retrieve current project context, updates, briefs, and task packets directly and record its result without moving provider credentials through chat.

1. Connect the endpoint from /dashboard/settings/mcp.
2. Complete OAuth and approve the intended account, project, and scopes.
3. List projects and check setup status.
4. Fetch one relevant update and its task packet.
5. Verify the local repository directly.
6. Record the update outcome after investigation or implementation.

See MCP for the exact tool sequence.

Export to Linear

Use Linear when the brief should become visible downstream work for a team. Export is explicit and creates one issue from one generated brief; it is not a broad or automatic sync.

See Linear for setup and duplicate-prevention behavior.

Agent Handoff Checklist

Before an agent starts, confirm the packet includes:

• the product unlock and why it matters;
• current source and repo evidence;
• unresolved questions;
• concrete starting files or product surfaces when known;
• the smallest safe action;
• explicit non-goals and trust boundaries;
• acceptance checks;
• the expected outcome to record.

What ShipFoundry Does

• Routes an opportunity into the output lane that matches its evidence and uncertainty.
• Produces bounded context for a user-owned agent.
• Supports copy/paste, MCP retrieval, and explicit Linear export.
• Keeps GitHub evidence read-only.
• Preserves uncertainty instead of presenting every match as implementation-ready.

What ShipFoundry Does Not Do

• It does not automatically implement updates.
• It does not grant a coding agent permission to modify every connected repository.
• It does not replace local repository inspection, tests, code review, or deployment approval.
• It does not expose GitHub, Linear, billing, or AI-provider credentials to the MCP client.
• It does not export every match to Linear.

Troubleshooting

If the packet is too broad, use an Engineering Spike and narrow the first inspection target.

If MCP returns no task packet, the update may not be grounded enough for an agent workflow. Review its trust state and missing evidence.

If the agent finds contradictory repo evidence, stop implementation, report the finding, and record the outcome so future matching can improve.

If Linear already contains the brief, use the stored export instead of creating a duplicate.

Related Pages

• Brief anatomy and output lanes: Investigation Briefs
• Safety states: Trust And Readiness
• Live agent connection: MCP
• Per-match repo proof: GitHub Repo Evidence
• Concrete artifact: Sample Engineering Spike
• Product CTA: Start 7-Day Trial

Was This Page Useful?

Send a short note if this page is unclear, overclaims a capability, or is missing the proof you expected.

Send Docs Feedback for Agent Workflows