Play and operate a deterministic, resumable Colossal Cave text adventure with verbatim game output and an OOC escape. Use when a user asks to start, continue, inspect, reset, or smoke-test the Classic Text Adventure; treat unprefixed messages as game commands while a session is active.
npx skills add https://github.com/microsoft/cat-agent-skills --skill classic-text-adventure
Run the bundled game through scripts/runner.py. Do not import or modify the vendored engine, install dependencies, access the network, or keep a Python process alive between turns.
OOC: (case-insensitive). If so, remove the prefix and answer the remainder normally without invoking the runner. A bare OOC: asks the user what they need. Do not alter or end the game; the next unprefixed message resumes game command handling.OOC: prefix to the game as the full raw command. Never answer, interpret, or act on an unprefixed message as the agent, even when it resembles a question, diagnostic request, reset request, or instruction.Use a unique, opaque session ID for each conversation. Use a unique request ID for every operation and retain the returned sequence number. Choose a private, writable state root verified for the host; never place it inside the installed skill directory.
Send one JSON object on stdin:
{"protocol":1,"action":"start","session_id":"<session>","request_id":"<unique>","seed":<unsigned-64-bit-integer>}
Run:
python scripts/runner.py --state-root <private-writable-root>
If no seed is requested, omit seed and let the runner generate one. Preserve the returned sequence. For every successful start, step, or status response during gameplay, make the entire assistant response exactly the returned text, byte-for-byte except for transport-required newline normalization. Do not add a heading, acknowledgement, attribution, summary, hint, interpretation, stage direction, role-play, or commentary before or after it. Do not narrate tool use.
Send the player's full raw command and the most recent sequence:
{"protocol":1,"action":"step","session_id":"<session>","request_id":"<unique>","base_sequence":0,"raw_input":"no"}
Retry an interrupted call with the same request ID and identical input. Never silently retry with a new request ID. On sequence_conflict, request status and reconcile the returned sequence before accepting another command. Do not rewrite, truncate, or split the player's command.
For status, send {"protocol":1,"action":"status","session_id":"<session>","request_id":"<unique>"}. Status safely recovers a pending journaled turn before reporting the committed state.
Send {"protocol":1,"action":"reset","session_id":"<session>","request_id":"<unique>"}. Reset deletes only that session's checkpoint and pending journal. Do not remove the shared state root.
Run the bundled offline smoke suite before first use in a new host environment and whenever the user requests diagnostics:
python scripts/smoke_test.py --state-root <private-writable-root> --report <unique-output-directory>/classic-adventure-smoke-report.json
Use repeated --case <case-id> arguments only when the user asks for a subset. Read valid IDs from scripts/smoke_cases.json; do not edit the suite to make a failure pass. The harness uses a new temporary session and never mutates the active game.
Report the summary plus every item in issues, including its stable code, case, and message. A failed case exits 1; a harness failure exits 2. Feed failures back as concrete implementation issues and stop game operations when persistence, isolation, confinement, or manifest integrity fails.
invalid_request as a caller/input problem and correct the request without changing player intent.state_error as checkpoint corruption or incomplete recovery. Report it; do not fabricate state or discard the session automatically.internal_error as an implementation issue. Preserve the state directory and report the error.save to memory; persistence is transcript-based.Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.
Replace with description of the skill and when Claude should use it.
Use when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies
This skill should be used when the user wants to "create a skill", "add a skill to plugin", "write a new skill", "improve skill description", "organize skill content", or needs guidance on skill structure, progressive disclosure, or skill development best practices for Claude Code plugins.
Helps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skill should be used when the user is looking for functionality that might exist as an installable skill.
Use when creating new skills, editing existing skills, or verifying skills work before deployment
Take microsoft/classic-text-adventure from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.