feat(coder/agent-relay-cursor): export the cloud session id

[matifali]

Summary

• Export session_id from the Cursor Agent Relay module. It is the agent_relay_session_id Agent Relay already stamps, and it is empty when a person creates the workspace by hand.
• When that id is set, the module adds two external apps, using Cursor's names for the clients:
  • Open in Cursor Web at https://cursor.com/agents/<id>
  • Open in Cursor Desktop at cursor://anysphere.cursor-deeplink/background-agent?bcId=<id>

Test plan

• A manual plan has an empty session_id and creates neither app (main.tftest.hcl)
• Dogfood agent-relay-template-docker pins this commit and a Cursor dispatch shows both buttons

[github-actions]

Module Scorecard Check

coder/agent-relay-cursor: 71 → 49

⚠️ Score regression: 71 → 49 (-22). Check the drilldown for which criteria dropped.

Theme	Before	After
Presentation & Onboarding	5 / 17	6 / 17
Integration	20 / 25	5 / 25
Credential Hygiene	20 / 20	12 / 20
Restricted-Environment	10 / 20	12 / 20
Engineering Quality	10 / 10	10 / 10
Overall	71 / 100	49 / 100

Full scorecard for this PR

Presentation & Onboarding	Agent Integration	Credential Hygiene	Restricted-Environment Readiness	Engineering Quality	Overall
6 / 17	5 / 25	12 / 20	12 / 20	10 / 10	49 / 100

Drilldown

Presentation & Onboarding — 6 / 17

Criterion	Max	Score	Notes
Configuration-mode examples	12	6	Only one code example is shown (install_cli). Other modes — computer_use, cli_binary override, repo-scoped pools, credential-kind switching — are described only in prose/tables (e.g. the insecure_shared_token credential-kind table), with no worked example for each mode.
Visual preview	5	0	No image, GIF, or video anywhere in the README.

Agent Integration — 5 / 25

Criterion	Max	Score	Notes
AI governance	10	0	No mention of Coder AI Gateway or Agent Firewall anywhere; the module integrates with "Agent Relay," a distinct dispatch mechanism, not AI Gateway/Agent Firewall.
Dashboard entry point	5	5	coder_app.cursor_web and coder_app.cursor_desktop are implemented and documented ("Open in Cursor Web"/"Open in Cursor Desktop" section), conditioned on agent_relay_session_id.
Session continuity	5	0	No documented resume/reconnect support or persistent session manager (tmux/screen). The module documents idle-release timers and worker restart/liveness bookkeeping, but this governs workspace reaping, not resuming an interrupted agent session.
Managed configuration	5	0	No documented MCP, settings, policy, or workdir management by this module (repo cloning/SCM credentials are explicitly pushed to the template, not handled here).

Credential Hygiene — 12 / 20

Criterion	Max	Score	Notes
Secrets marked sensitive	16	8	agent_relay_credential uses styling.mask_input = true and ephemeral = true, not a literal sensitive = true. README examples show no inline/plaintext secrets, and the "Worker credential" section explains masking and non-persistence in detail, but the literal sensitive-marking mechanism required by the rubric is absent.
Non-hardcoded auth path	4	4	README's "Worker credential" section documents the service-account key never leaving Agent Relay; a per-user, short-lived sub-token is exchanged automatically and injected via env var — no raw key pasted into the template.

Restricted-Environment Readiness — 12 / 20

Criterion	Max	Score	Notes
Mirrorable artifact source	5	0	install.sh.tftpl hardcodes curl https://cursor.com/install -fsSL | bash. No module variable (only cli_binary, install_cli, state_file, log_file, serving_log_pattern, computer_use exist) overrides this download URL.
Bring-your-own binary	10	10	install_cli = false fully skips download; README documents this explicitly ("Bake it into the image and set this to false for faster workspaces," plus the Requirements section), and tests confirm a pre-baked ~/.local/bin/agent is found.
Egress transparency	3	0	No dedicated README section enumerating install/runtime endpoints (e.g. cursor.com) or notes for restricted/air-gapped environments; the domain only appears in script comments, not the README.
Runs without sudo	2	2	install.sh.tftpl, start.sh.tftpl, and status.sh.tftpl contain no sudo invocations anywhere; scripts run entirely as the workspace user.

Engineering Quality — 10 / 10

Criterion	Max	Score	Notes
Input quality	6	6	Clear, thorough descriptions for every variable; sensible defaults (install_cli = true, computer_use = false); validation blocks on cli_binary (command-safe regex) and serving_log_pattern (non-empty).
Test coverage	4	4	main.tftest.hcl covers the parameter contract, script wiring, credential-kind branching, and injection-safety; main.test.ts runs true end-to-end container tests covering install/start lifecycle, restart semantics, liveness detection, and shell-injection safety.

Overall — 49 / 100

Raw 45 / 92 → round(45 / 92 × 100) = 49

Tip

You can run this locally by telling your agent: "review this module against .github/scorecard/SCORECARD.md".

---

Scored against SCORECARD.md with solstice-1. Language-model scores are advisory.

[Emyrk]

Oh this is cool

[matifali]

Tested this in the Dogfood environment.