The terminal's System screen says how the forge is: the version and commit it runs, draining or not, the rollout and its incidents, as MCP `health` does #7304

The terminal's System screen now opens with the forge's own state, above the machine lines it already showed. It reads readHealth(session.app, session.user), the same answer MCP health gives that user.

• First line, always: forge <version> · <7-character commit> · serving (or draining). When nothing says which commit is running, it shows no commit.
• Second line, only when there is something to say: whether the rollout is blocked or held, or the platform's view is stale, plus how many other incidents there are and what each says (key: message). It is cut to the dialog's width with …. stuck and held are already in the rollout's words, so they are not counted again as incidents. When all is well, nothing more is drawn.
• Who sees it: the same as MCP. The command is still offered only to forge admins who are people. forgeLines (src/tui/screens/system.ts) takes any readHealth answer, and a signed-out caller's answer has no incidents in it, so it never gets the second line.

Tests: test/tui-system.test.ts sets a fake version and commit and seeds incidents in a temporary FORGE_GUARD_DIR. It checks:

• With no incidents there is exactly one line, both from forgeLines and in the opened screen.
• One seeded incident adds exactly one line.
• A blocked rollout and an incident share that one line, and a narrow width cuts it with ….
• A signed-out caller gets one line even with an incident seeded.

Checks run:

• Passed: bun x tsc --noEmit, test/mcp-size.test.ts, the design lint, and seven nearby suites (system, telemetry, resilience, tui screens, size, access).
• Not run: scripts/fast.ts named 160 files, which it estimated at about 13 minutes, so I left the full run to the forge's check on push.

Chosen not to do:

• No new MCP tool and no change to the health tool.
• For an admin, readHealth also gathers the forge's vitals when the command opens. I kept that so the screen reads exactly what MCP returns, rather than calling forgeHealth directly to save the work.

To try it: sign in to the terminal as a forge admin and open the System command. To see the second line, point FORGE_GUARD_DIR at a folder whose status.json is {"incidents":{"down":"…"}}.

🤖 Generated with Claude Code

Closes #7296

The terminal's System screen now opens with the forge's own state, above the machine lines it already showed. It reads readHealth(session.app, session.user), the same answer MCP health gives that user.

• First line, always: forge <version> · <7-character commit> · serving (or draining). When nothing says which commit is running, it shows no commit.
• Second line, only when there is something to say: whether the rollout is blocked or held, or the platform's view is stale, plus how many other incidents there are and what each says (key: message). It is cut to the dialog's width with …. stuck and held are already in the rollout's words, so they are not counted again as incidents. When all is well, nothing more is drawn.
• Who sees it: the same as MCP. The command is still offered only to forge admins who are people. forgeLines (src/tui/screens/system.ts) takes any readHealth answer, and a signed-out caller's answer has no incidents in it, so it never gets the second line.

Tests: test/tui-system.test.ts sets a fake version and commit and seeds incidents in a temporary FORGE_GUARD_DIR. It checks:

• With no incidents there is exactly one line, both from forgeLines and in the opened screen.
• One seeded incident adds exactly one line.
• A blocked rollout and an incident share that one line, and a narrow width cuts it with ….
• A signed-out caller gets one line even with an incident seeded.

Checks run:

• Passed: bun x tsc --noEmit, test/mcp-size.test.ts, the design lint, and seven nearby suites (system, telemetry, resilience, tui screens, size, access).
• Not run: scripts/fast.ts named 160 files, which it estimated at about 13 minutes, so I left the full run to the forge's check on push.

Chosen not to do:

• No new MCP tool and no change to the health tool.
• For an admin, readHealth also gathers the forge's vitals when the command opens. I kept that so the screen reads exactly what MCP returns, rather than calling forgeHealth directly to save the work.

To try it: sign in to the terminal as a forge admin and open the System command. To see the second line, point FORGE_GUARD_DIR at a folder whose status.json is {"incidents":{"down":"…"}}.

🤖 Generated with Claude Code