sts2-ironclad-agent
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| STS2_AGENT_CONFIG | Yes | Path to the local config.toml file containing game, KEV, and Luna service addresses. | |
| STS2_LUNA_API_KEY | No | API key for Luna (OpenAI-compatible endpoint). Either this or SSH key file settings must be provided if Luna credentials are used. | |
| STS2_LUNA_KEY_SSH_TARGET | No | SSH target for fetching a Luna key file. Must be set together with STS2_LUNA_KEY_REMOTE_PATH. | |
| STS2_LUNA_KEY_REMOTE_PATH | No | Remote path of the Luna key file. Must be set together with STS2_LUNA_KEY_SSH_TARGET. |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": false
} |
| prompts | {
"listChanged": false
} |
| resources | {
"subscribe": false,
"listChanged": false
} |
| experimental | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| preflightA | Read-only check of the local game bridge, local game version, Kev model list and Luna key readiness. |
| get_game_snapshotA | Read state only from bridge schema 2, which advertises passive snapshots; refuse older bridges before requesting state. |
| get_legal_actionsA | Read and validate the action candidates exposed by the current game screen. |
| plan_map_routeA | Ask Luna to plan a full Ironclad route to the boss. Planning does not move the character. |
| recommend_next_actionC | Create one validated proposal: KEV for tactical choices, Luna for strategy, deterministic for forced or unique actions. |
| execute_proposalC | Re-read the game and submit exactly one still-legal proposal. Dry-run is the default; no retries on timeout. |
| step_onceC | Run one routed decision and, only when enabled in local config, submit at most one validated game action. |
| autoplayB | Immediately continue Ironclad decisions and legal actions for up to 20 steps; stops on uncertainty or an optional Act 1 boss clear. |
| start_new_ironclad_run_stepB | Start a fresh Ironclad run with one advertised menu action per call; never switch profiles or overwrite a continue run. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 9 tools
Most tools have clear, distinct roles across the pipeline: readiness, snapshotting, legal actions, planning, recommendation, execution, stepping, autoplay, and new-run setup. The only mild boundary ambiguity is between execute_proposal and step_once, both of which can submit a single action, but their descriptions distinguish lower-level proposal submission from routed decision execution.
All names are readable snake_case, but they do not follow a single pattern. get_game_snapshot and get_legal_actions use get_, while plan_map_route, recommend_next_action, and execute_proposal use verb_noun, and preflight, step_once, and autoplay are one-off or compound names.
Nine tools is a well-scoped size for an autonomous game agent. Each tool supports a distinct stage of the play loop without excessive redundancy or unnecessary surface area.
The tool set covers the full gameplay automation loop: preflight checks, state reads, legal-action discovery, path planning, decision recommendation, proposal execution, single-step control, bounded autoplay, and starting a new run. Minor gaps exist around explicit stop/pause or richer configuration control, but the existing tools handle those through local config and autoplay stopping conditions.