GitHub Copilot CLI hooks
Download and install jq before adding this hook, and make sure it is on your PATH. The bash command uses it to escape stagelint's output as JSON.
If stagelint is installed in your project or through a tool manager, invoke it the way that tool does, such as npx stagelint, uv run stagelint or mise exec -- stagelint.
Configure the hook in .github/hooks/stagelint.json:
{
"version": 1,
"hooks": {
"agentStop": [
{
"type": "command",
"bash": "out=$(stagelint --unstaged --quiet 2>&1) || printf '%s' \"$out\" | jq -Rs '{decision:\"block\",reason:.}'",
"powershell": "$out = stagelint --unstaged --quiet 2>&1 | Out-String; if ($LASTEXITCODE -ne 0) { @{decision='block'; reason=$out} | ConvertTo-Json -Compress }"
}
]
}
}What each part does
agentStop fires when the main agent finishes a turn.
--unstaged runs your commands against the working tree rather than the index, because the agent's edits are not staged yet. Nothing is stashed and nothing is staged. --quiet prints only the output of failed commands, which keeps the model's context small.
decision: "block" on stdout is the only route back to the model, and forces another agent turn using reason as the prompt. No exit code reaches it: exit code 2 shows stderr to you as a warning, any other code is only logged, and the turn ends either way.
After 8 consecutive blocks Copilot ends the turn anyway, so a failure the agent cannot fix does not loop forever.
If every task in your config fixes what it finds, the agent does not need to hear about it, and a plain "command": "stagelint --unstaged --quiet" is enough.