GitHub Copilot CLI + tman

The hook can allow, ask, or deny — but not rewrite. tman still decides whether the command runs; getting it supervised takes a shim or a denial the model can act on.

vendor: GitHub
integration: gates
hook event: preToolUse
config: .github/hooks/copilot-cli-policy.json

Copilot CLI exposes three hook events — sessionStart, userPromptSubmitted, and preToolUse — and the pre-tool one is explicitly a policy gate: it can allow execution by doing nothing, or deny it by returning a structured response. Commands cannot be rewritten. The policy file lives in the repository, which makes this the easiest of the gate-tier integrations to apply to a whole team at once.

1. Install tman and adopt the project

Everything below assumes the binary is on your PATH and the repository has a .tman.kdl. That file is what scopes caps to this project, and it is also the signal several of the integrations test for before they do anything.

shell

npm install -g @standardbeagle/tman   # or the install.sh one-liner

cd your-project
tman init --shims --gitignore

tman init writes a defaults block and an alias for each test or build command it can detect. Commands it cannot detect are left commented out rather than stubbed, so an alias you never filled in fails loudly instead of exiting 0 having run nothing.

2. Commit the hooks policy

Because .github/hooks/copilot-cli-policy.json is a checked-in file, adopting tman and committing this policy in the same change means every clone of the repository gets the same supervision. Note the payload quirk: toolArgs arrives as a JSON string and has to be parsed before you can read the command out of it.

.github/hooks/copilot-cli-policy.json

{
  "version": 1,
  "hooks": {
    "preToolUse": [
      {
        "type": "command",
        "bash": "~/.config/tman-gate.sh",
        "timeoutSec": 10
      }
    ]
  }
}

Turn the rewrite into an actionable denial

Because this hook cannot replace the command, the useful move is to refuse the unsupervised one and hand the model the supervised string to re-issue. tman already knows which commands deserve that and what the replacement should be, so the adapter is a few lines around tman hook pretooluse rather than a second copy of the classifier.

~/.config/tman-gate.sh — chmod +x

#!/usr/bin/env bash
# tman-gate.sh — deny an unsupervised test/build command, naming its supervised form.
set -euo pipefail
payload=$(cat)

# (.toolArgs | if type == "string" then fromjson else . end).command // empty
command=$(jq -r '(.toolArgs | if type == "string" then fromjson else . end).command // empty' <<<"$payload")
[ -n "$command" ] && [ "$command" != "null" ] || exit 0

verdict=$(jq -nc --arg c "$command" --arg d "$PWD" \
  '{tool_name:"Bash", cwd:$d, tool_input:{command:$c}}' | tman hook pretooluse)

# tman stays silent on anything it will not supervise. So does this.
[ -n "$verdict" ] || exit 0
supervised=$(jq -r '.hookSpecificOutput.updatedInput.command // empty' <<<"$verdict")
[ -n "$supervised" ] || exit 0

jq -nc --arg s "$supervised" '{
  permissionDecision: "deny",
  permissionDecisionReason: ("Run it supervised instead: " + $s)
}'

Confirm the payload field before you trust it. Hook payload shapes move between versions, and a jq path that no longer matches yields an empty command and a script that silently allows everything — a false green. Replace the script with cat > /tmp/hook-payload.json for one run, read the file, then wire the real adapter against the field name you actually saw.

Copilot's denial shape is {"permissionDecision":"deny","permissionDecisionReason":"…"}, which is exactly what the adapter above emits — no change needed to its final jq.

3. Custom instructions

Copilot CLI reads AGENTS.md, and also .github/copilot-instructions.md, .github/instructions/**.instructions.md, CLAUDE.md, and GEMINI.md. If you already keep a shared AGENTS.md for other agents, one copy covers Copilot too.

Tell the model, in AGENTS.md

Shims and hooks catch the mechanical cases. The remaining gap is the model deciding to do something clever — running the suite through a subshell, or with an inline environment assignment, both of which tman deliberately refuses to rewrite because prefixing a string it did not parse changes what runs. A standing instruction in AGENTS.md closes that gap:

AGENTS.md

## Running tests and builds

Run every test, build, lint, and typecheck command through tman — never bare.
Prefer the repo-root shims (`./test`, `./lint`) when they exist; they carry
this project's `.tman.kdl` caps. Otherwise prefix explicitly:

    tman run -- <command>

Never widen the caps from the command line. Exit 124 is a timeout, 125 a stall,
126 a resource cull — report those, do not retry with looser limits.

Verify it is actually on

A supervision integration that silently does nothing is worse than none, because it stops you looking. Two checks, both of which fail visibly:

shell

# 1. the classifier agrees this command is worth supervising
echo '{"tool_name":"Bash","cwd":"'"$PWD"'","tool_input":{"command":"npm test"}}' \
  | tman hook pretooluse

# 2. a real run shows up while it is in flight
tman run --name smoke -- sleep 30 &
tman list

The first prints a JSON object naming the rewritten command. Empty output means tman decided to stay out of the way — no .tman.kdl above cwd, a command it does not classify as test or build work, or a shell string it refused to parse. The second must list the run; if tman list is empty, nothing is being supervised.

Caps worth setting for agent-driven work

An agent re-runs the same command far more often than a human does, and it does it while you are not watching. That changes which caps earn their keep:

cap	why it matters more under an agent
`--name`	The agent will start the suite again before the last one finished. A dedup lock refuses the duplicate instead of running two copies against the same port or database.
`--max-parallel 2`	The default, and the right one for most repos. Excess runs queue on a slot file rather than stampeding your cores.
`--max-time`	The only cap that bounds a run which is genuinely working but will never finish — a wait on a peer that never answers looks identical to healthy IO.
`--stall 30m`	The built-in. A hang backstop, not a runtime budget: it should sit well above the longest quiet stretch your slowest command legitimately has.
`--max-mem`	Opt-in, and worth it where a leak mode exists — worker pools, browser fleets, long-lived daemons. Off by default because builds are supposed to be greedy.

Per-command starting points live in the tuning guides — one page per test and lint tool, with the reasoning behind each number rather than just the number.

Frequently asked

Can a Copilot CLI hook modify the command it inspects?

No. The preToolUse hook allows execution by doing nothing, or denies it by returning {"permissionDecision":"deny","permissionDecisionReason":"…"}. There is no field for a replacement command, so supervision comes from PATH shims plus a denial naming the supervised form.

Why does my hook see toolArgs as a string?

That is the documented shape: `toolArgs` arrives JSON-encoded inside the payload and needs a second parse. In jq, `(.toolArgs | if type == "string" then fromjson else . end).command` handles both forms, which keeps the adapter working if the encoding changes.

Other agents

Sources

Vendor documentation this guide is written against — hook surfaces move, so check yours if something below does not match what you see.