harness-commands

Files (2)

• LIMITATIONS.md
• SKILL.md

SKILL.md (first lines)

# Know your own harness before you run or mail a native command

footnote drives several coding CLIs (claude, codex, opencode, agy, pi, cursor-agent, grok). Each has its own native slash commands. The measured roster, each command's risk class, and a one-line when-to-use live in the capability table. `fno-agents commands` renders your row of it.

## Run this first

```bash
fno-agents commands
```

Bare form detects your own harness from the session env. On a refusal, pass the name: `fno-agents commands claude`. `--json` for machines.

Read the roster before you reach for a native command you have not run in this harness. The risk column is the mail guard's own vocabulary: `safe`, `context-destroying` (clears or swaps the session's context), `session-ending` (exits or logs out).

## The two raw-mail rules

`fno agents mail send --raw` runs a command on another session, as user-shaped text. Two rules:

1. `--raw` is commands only, never a message. A message goes wrapped (the plain send wraps it, and the sender stays visible).
2. A `session-ending` or `context-destroying` command is refused by the mail guard unless the send names it: `--ack-command-risk <command>`. The refusal names the risk class and the remedy. An acked send still carries the command's full effect. The ack is an acknowledgment, not a safety net.

## Pitfalls

Harness-specific failure modes (symptom, cause, fix) are tabled per harness in `docs/HARNESSES.md`. Read your harness's table before debugging anything harness-specific.

## Honesty

The table says which rows are vendor-measured and which are name-derived, right beside the data. Never claim a command does what the table does not say.

## Known Limitations and Deferred Work

- The render is only as current as the table's measurement date. A harness newer than its row drifts it. See [LIMITATIONS.md](LIMITATIONS.md).

From bllshttng/fno, Apache-2.0.

More from bllshttng/fno