iterating

The iterating skill provides a structured methodology for managing complex, long-running agentic tasks that exceed the token limits or time constraints of a single conversation session. By implementing a formal 'Checkpoint Pattern,' it ensures that project knowledge is not lost, allowing users and agents to resume work seamlessly across different sessions. This skill is specifically designed for developers and researchers who need to maintain state during involved processes such as refactoring, feature development, debugging, or complex research initiatives where context accumulation is critical.

• Implements a standardized WorkLog format with version control, status tracking (in_progress, blocked, needs_review, completed), and priority levels (HIGH, MED, LOW).

• Automatically detects the host environment (Claude Code CLI, Claude Desktop, Web, or remote sessions) to determine the most effective persistence method for Work Logs.

• Enforces a stop-and-verify workflow: the agent completes one high-priority item, updates the log, and pauses, preventing the 'waterfall' behavior that often leads to context loss or token exhaustion.

• Provides built-in templates for project documentation, including decision logs, file-specific references (using relative paths and line ranges), and progress metrics.

• Supports manual parsing of past Work Logs to reconstruct task state when starting new conversations.

• Use this skill when working on projects that require multiple sessions; it is ideal for debugging legacy code, large-scale feature implementation, or deep technical research.

• Always save the Work Log to your project knowledge base after the agent generates or updates it to ensure maximum portability.

• When prompted to 'continue,' the agent will automatically parse the latest WorkLog version, identify the next priority task, and execute precisely one item before stopping again for review.

• Ensure all file references use relative paths from the project root to maintain compatibility across different compute environments.

• The 'STOP' step is mandatory: do not attempt to bypass it, as it is the primary mechanism for managing token usage and ensuring high-quality, iterative progress.