name	cortexfs
description	This skill should be used when working on the dependency-free agent.sh CLI as a CortexFS frontend, including AI calls through the filesystem ABI, MCP stdio, and cluster task submission.
version	0.2.0

cortexfs

Core identity: this is CortexFS infrastructure exposed through a tiny shell agent. The skill is about the CortexFS ABI, not about terminal execution.

Rules:

Keep agent.sh below 200 lines.
Do not add dependencies.
Treat CortexFS as the execution plane; do not bypass it for AI calls or cluster tasks.
MCP stdio messages are JSON-RPC lines.
Terminal execution is only one explicit MCP tool, terminal.run; do not let it define the project.
Cluster work must be explicit through cluster.submit.

Test:

bash -n agent.sh
./agent.sh help
./agent.sh run 'printf "%s\n" ok'
printf '%s\n' '{"jsonrpc":"2.0","id":"1","method":"tools/list"}' | ./agent.sh mcp

name	cortexfs
description	This skill should be used when working on the dependency-free agent.sh CLI as a CortexFS frontend, including AI calls through the filesystem ABI, MCP stdio, and cluster task submission.
version	0.2.0

Core identity: this is CortexFS infrastructure exposed through a tiny shell agent. The skill is about the CortexFS ABI, not about terminal execution.

Rules:

Keep agent.sh below 200 lines.
Do not add dependencies.
Treat CortexFS as the execution plane; do not bypass it for AI calls or cluster tasks.
MCP stdio messages are JSON-RPC lines.
Terminal execution is only one explicit MCP tool, terminal.run; do not let it define the project.
Cluster work must be explicit through cluster.submit.

Test:

bash -n agent.sh
./agent.sh help
./agent.sh run 'printf "%s\n" ok'
printf '%s\n' '{"jsonrpc":"2.0","id":"1","method":"tools/list"}' | ./agent.sh mcp