Completion guard that prevents premature stopping. Enforces an explicit TASKMASTER_DONE::<session_id> signal before allowing the session to end. This skill is enforced by a Stop hook — it activates automatically whenever Claude tries to stop. Before stopping, verify: (1) the user's stated goal is fully achieved (yes/no, not "partially"), (2) every discrete request is FULLY done, (3) all tasks are completed, (4) all verification steps executed and passing, (5) no errors, TODOs, or loose ends remain. Do not narrate remaining work — execute it. Progress is not completion. Use this skill proactively whenever working on multi-step tasks, complex features, or any task where premature stopping would waste the user's time.
From claude-taskmasternpx claudepluginhub hex/claude-marketplace --plugin claude-taskmasterThis skill uses the workspace's default tool permissions.
Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Implements structured self-debugging workflow for AI agent failures: capture errors, diagnose patterns like loops or context overflow, apply contained recoveries, and generate introspection reports.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
A session is complete ONLY when you emit this exact line on its own line:
TASKMASTER_DONE::<session_id>
If you stop without this signal, the Stop hook will block you and you must continue working.
Before emitting the done signal, verify ALL of the following in order:
Answer these questions explicitly in your response:
"Diminishing returns", "no single dominant hotspot", "would require broader architectural changes", or any variation of "I made good progress" are NOT valid reasons to stop. These are rationalizations.
DO NOT NARRATE — EXECUTE. If incomplete work remains, your ONLY job is to DO that work right now. Do not describe what remains, analyze its complexity, list dependencies, or ask for permission. Open files, write code, run commands, fix bugs. Act.
PROGRESS IS NOT COMPLETION. "I improved X from 1345ms to 866ms" does not satisfy a goal of "<500ms". Describing remaining work with phrases like "would require deeper analysis" is narrating — not doing. If the goal is not met, keep working.
HONESTY CHECK. Before marking anything as "not possible" or "skipped", ask yourself: did you actually TRY? "I can't do X" is almost never true — what you mean is "I haven't tried X yet." Attempt it first.
USER PRIORITY. The user's latest instructions always take priority. If the user said to stop, move on, or skip something, respect that — do not force completion of work the user no longer wants.
TASKMASTER_MAX (env var, default 0): Max stop-blocks before allowing
stop. 0 means unlimited — hook keeps firing until done signal is emitted.