Function buildSessionWorkspaceHint

cli/agent_mode.go:3229–3267

model how to use the per-session scratch dir and how to recover data from truncated tool results. Returns an empty string when the session workspace hasn't been initialized (shouldn't happen during normal startup).

()

// truncated tool results. Returns an empty string when the session workspace
// hasn't been initialized (shouldn't happen during normal startup).
func buildSessionWorkspaceHint() string {
	ws := agent.GetSessionWorkspace()
	if ws == nil {
		return ""
	}
	return "\n\n## SESSION WORKSPACE & LARGE OUTPUTS\n" +
		"\n" +
		"You have an isolated scratch directory for this session, exposed via the\n" +
		"environment variable `CHATCLI_AGENT_TMPDIR` (current value: `" + ws.ScratchDir + "`).\n" +
		"Both read and write are ALLOWED in this directory and in its subtree —\n" +
		"use it whenever you need to:\n" +
		"- stage a temporary shell script before exec'ing it;\n" +
		"- persist an intermediate artifact between tool calls;\n" +
		"- avoid polluting the project tree with one-off files.\n" +
		"\n" +
		"Example: `exec { \"cmd\": \"cat > $CHATCLI_AGENT_TMPDIR/patch.sh <<'EOF' ...\" }`.\n" +
		"\n" +
		"### Truncated tool outputs\n" +
		"\n" +
		"When a tool result is large, ChatCLI automatically truncates it inline\n" +
		"and saves the FULL output to a file in this session. You will see a\n" +
		"marker like:\n" +
		"\n" +
		"    ... [N chars omitted — full output saved to /tmp/chatcli-agent-XXX/tool-results/budget_xxx.txt]\n" +
		"    ... [full output saved to /tmp/chatcli-agent-XXX/tool-results/result_XXX.txt — N bytes total]\n" +
		"\n" +
		"When you see such a marker and the omitted portion matters, use the\n" +
		"`read_file` tool with `start` / `end` line numbers to examine specific\n" +
		"ranges of the saved file — do NOT re-run the original tool call.\n" +
		"\n" +
		"### Delegating heavy analysis\n" +
		"\n" +
		"For tasks that would otherwise flood your context with raw data (large\n" +
		"metrics endpoints, verbose logs, wide-scope searches), use the\n" +
		"`delegate_subagent` tool. The subagent runs with its OWN isolated\n" +
		"context window; only its final summary returns to you. Example use\n" +
		"cases: \"summarize memory hotspots from /metrics\", \"find all call\n" +
		"sites of func X across the repo\".\n"
}

// isDelegateInvocation reports whether an @coder JSON args payload is a
// delegate_subagent call rather than a normal engine subcommand.