You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A discussion dedicated to the Agent Relay Claude Code module. Share your thoughts, questions, and feedback here.
Module Scorecard
Presentation & Onboarding
Agent Integration
Credential Hygiene
Restricted-Environment Readiness
Engineering Quality
Overall
6 / 17
10 / 25
20 / 20
12 / 20
10 / 10
63 / 100
Drilldown
Presentation & Onboarding — 6 / 17
Criterion
Max
Score
Notes
Configuration-mode examples
12
6
README shows the main install_cli=true path with a full code block, plus separate examples for additional_args, release_idle_session_min, and health_port. But the "baked-in image" mode (install_cli=false) only gets an inline comment, not a full worked example, and there isn't a clear "major mode" structure (e.g. provider choice) beyond this single download toggle.
Visual preview
5
0
No image, GIF, or video anywhere in the README.
Credential Hygiene — 20 / 20
Criterion
Max
Score
Notes
Secrets marked sensitive
16
16
The module is explicitly designed so the work-order JWT (SELF_HOSTED_RUNNER_ENVIRONMENT_SECRET) never becomes a module variable or coder_env/parameter — it's injected directly by the agent from an ephemeral workspace secret, verified by a dedicated test ("sets no coder_env for the work order's variable"). No example in the README contains inline secrets.
Non-hardcoded auth path
4
4
README's "Requirements" section explicitly documents the ephemeral-workspace-secret delivery path as the only mechanism, replacing the old agent_relay_credential build parameter.
Restricted-Environment Readiness — 12 / 20
Criterion
Max
Score
Notes
Mirrorable artifact source
5
0
install.sh.tftpl hardcodes curl https://claude.ai/install.sh -fsSL | bash. No variable overrides this URL; cli_binary only overrides the binary path, not the download source.
Bring-your-own binary
10
10
install_cli = false disables the download entirely and documents expecting cli_binary already in the image ("Bake the CLI into the image and set this to false for faster workspaces"); exercised by multiple tests (finds a bring-your-own CLI…, skips the download…).
Egress transparency
3
0
No dedicated network/egress/air-gapped section. The claude.ai/install.sh endpoint appears only in the script source, not documented anywhere in the README text itself.
Runs without sudo
2
2
install.sh.tftpl, start.sh.tftpl, and cleanup.sh.tftpl never invoke sudo; all operations (curl install, process management, file writes) use user-owned paths like $HOME/.local/bin and $HOME/.coder-modules.
Engineering Quality — 10 / 10
Criterion
Max
Score
Notes
Input quality
6
6
Every variable has a clear description and sensible default; cli_binary, exit_if_unused_min, release_idle_session_min, health_port, and additional_args all carry validation blocks with precise bounds and anti-footgun checks (e.g. rejecting reserved flags, shell metacharacters).
Test coverage
4
4
Extensive .tftest.hcl covers parameter contract, env wiring, and rendered-script logic; main.test.ts runs true end-to-end container tests covering lifecycle states (idle/working/done/failed/orphaned), cleanup races, stale PID handling, and argument escaping.
Agent Integration — 10 / 25
Criterion
Max
Score
Notes
AI governance
10
0
No mention of Coder AI Gateway or Agent Firewall anywhere in the README; "Agent Relay" is a separate dispatch/pool mechanism, not the AI Gateway/Agent Firewall governance referenced by this criterion.
Dashboard entry point
5
0
No coder_app resource or example; README explicitly instructs against exposing the health port via coder_app ("Do not publish health_port through a coder_app…").
Session continuity
5
5
"Idle session release" section explicitly documents resuming: "The user's next message resumes the session on a fresh runner," backed by the release_idle_session_min mechanism.
Managed configuration
5
5
base_dir (workdir), additional_args, exit_if_unused_min, release_idle_session_min, and health_port are all documented, overridable runner policy/config knobs.
Overall — 63 / 100
Raw 58 / 92 → round(58 / 92 × 100) = 63
Scored against SCORECARD.md on 2026-10-29 with solstice-1.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
A discussion dedicated to the Agent Relay Claude Code module. Share your thoughts, questions, and feedback here.
Module Scorecard
Drilldown
Presentation & Onboarding — 6 / 17
install_cli=truepath with a full code block, plus separate examples foradditional_args,release_idle_session_min, andhealth_port. But the "baked-in image" mode (install_cli=false) only gets an inline comment, not a full worked example, and there isn't a clear "major mode" structure (e.g. provider choice) beyond this single download toggle.Credential Hygiene — 20 / 20
SELF_HOSTED_RUNNER_ENVIRONMENT_SECRET) never becomes a module variable orcoder_env/parameter — it's injected directly by the agent from an ephemeral workspace secret, verified by a dedicated test ("sets no coder_env for the work order's variable"). No example in the README contains inline secrets.agent_relay_credentialbuild parameter.Restricted-Environment Readiness — 12 / 20
install.sh.tftplhardcodescurl https://claude.ai/install.sh -fsSL | bash. No variable overrides this URL;cli_binaryonly overrides the binary path, not the download source.install_cli = falsedisables the download entirely and documents expectingcli_binaryalready in the image ("Bake the CLI into the image and set this to false for faster workspaces"); exercised by multiple tests (finds a bring-your-own CLI…,skips the download…).claude.ai/install.shendpoint appears only in the script source, not documented anywhere in the README text itself.install.sh.tftpl,start.sh.tftpl, andcleanup.sh.tftplnever invokesudo; all operations (curl install, process management, file writes) use user-owned paths like$HOME/.local/binand$HOME/.coder-modules.Engineering Quality — 10 / 10
cli_binary,exit_if_unused_min,release_idle_session_min,health_port, andadditional_argsall carryvalidationblocks with precise bounds and anti-footgun checks (e.g. rejecting reserved flags, shell metacharacters)..tftest.hclcovers parameter contract, env wiring, and rendered-script logic;main.test.tsruns true end-to-end container tests covering lifecycle states (idle/working/done/failed/orphaned), cleanup races, stale PID handling, and argument escaping.Agent Integration — 10 / 25
coder_appresource or example; README explicitly instructs against exposing the health port viacoder_app("Do not publishhealth_portthrough acoder_app…").release_idle_session_minmechanism.base_dir(workdir),additional_args,exit_if_unused_min,release_idle_session_min, andhealth_portare all documented, overridable runner policy/config knobs.Overall — 63 / 100
Raw 58 / 92 → round(58 / 92 × 100) = 63
Scored against SCORECARD.md on 2026-10-29 with
solstice-1.All reactions