dkloop

name	dkloop
description	Run a prompt in a loop until the requested work is fully implemented and verified.

Skill: dkloop

Run a prompt in a loop until fully implemented. In-session equivalent of the dkloop CLI command.

When to Use

• When the user wants to ensure a task is 100% complete before stopping
• When the user invokes /dkloop <prompt>

Arguments

The argument is the prompt/task to execute. Example: /dkloop add input validation to all API endpoints

Steps

1. Activate the loop

Run this bash command to activate the stop hook loop and clean stale state:

source "${DOYAKEN_DIR:-$HOME/work/doyaken}/lib/common.sh"
mkdir -p "$DK_LOOP_DIR"
SESSION_ID="${DOYAKEN_SESSION_ID:-$(dk_session_id)}"
touch "$(dk_active_file "$SESSION_ID")"
rm -f "$(dk_loop_file "$SESSION_ID")" "$(dk_complete_file "$SESSION_ID")"
echo "Loop activated (session: ${SESSION_ID})"

This creates an .active signal file that the stop hook checks. The hook will prevent you from stopping until you write the .complete file.

2. Execute the prompt

Work on the user's prompt. Follow the same approach as any implementation task:

• Read relevant code before making changes
• Follow existing patterns and conventions
• Test your changes where possible

3. Self-audit before completing

Before signaling completion, critically review your work against the original prompt. The stop hook will guide you through a comprehensive audit when you try to stop, including:

1. Acceptance criteria extraction — list all requirements from the prompt (explicit and implied)
2. /dkreview --single-pass — run /dkreview --single-pass on your changes (code-change sessions only)
3. Multi-perspective inventory — 4-pass manual review (Logic & Correctness, Structure/Design/Documentation, Security, Holistic Consistency & Dependencies)
4. Self-reviewer agent — spawn the self-reviewer agent for an independent adversarial review (code-change sessions only)
5. Merged inventory and batch fix — combine all findings from /dkreview --single-pass, manual passes, and the agent; fix in severity order
6. /dkverify quality pipeline — format, lint, typecheck, test (code-change sessions only)
7. Evidence table — trace each requirement to specific file:line evidence in code and tests

If you find issues at any step, fix them and re-audit. Do NOT proceed to step 4 until the audit passes.

4. Signal completion

After the audit passes, stop. The Stop hook manages completion — it will provide the completion signal file path and promise string after enough quality audit passes. Do NOT write any .complete files or output promise strings on your own.

Notes

• The stop hook will block you from stopping and inject an audit prompt. Follow the audit instructions — the hook provides completion instructions after sufficient clean iterations.
• If the loop reaches max iterations (default 30), it will allow stopping as a safety net.
• The .active file is cleaned up automatically when the loop completes or reaches max iterations.