workflow-patterns

name	Workflow Patterns
description	Patterns for output formats, reflection checkpoints, agent references, and self-evaluation shared across agents and commands.
version	2.0.0

Provide standardized patterns for output formatting, workflow checkpoints, agent references, and self-evaluation shared across agents and commands. Standard agent output format with status criteria Include in agents for consistent output structure Standard analysis quality checkpoint for workflow phases Include in workflows for consistent quality gates Standard self-evaluation phase for agents producing reports Include in agents for consistent quality assessment Standard Serena initialization phase for commands Include at start of command workflows for Serena initialization Serena validation elements for reflection checkpoints Include in checkpoints before code modifications or after investigations Standard agent output format with status criteria { "status": "success|warning|error", "status_criteria": { "success": "All checks passed, confidence >= 80", "warning": "Minor issues OR confidence 60-79", "error": "Critical issues OR confidence less than 60" }, "confidence": 0, "summary": "Brief summary of results", "metrics": {}, "findings": [], "next_actions": [] } Standard status criteria for agent output format "status_criteria": { "success": "All checks passed, confidence >= 80", "warning": "Minor issues OR confidence 60-79", "error": "Critical issues OR confidence less than 60" } Standard analysis quality checkpoint for workflow phases Have I gathered sufficient evidence to proceed? Are there gaps in my understanding? If confidence less than 70, seek more evidence or ask user Standard Serena initialization phase for workflows Initialize Serena and check existing patterns Activate Serena project with activate_project Workflow guidance Step completed Check list_memories for relevant patterns Workflow guidance Step completed Load applicable memories with read_memory Workflow guidance Step completed Serena validation for investigation/analysis checkpoints think_about_collected_information After investigation phase completes Serena validation before code modification operations think_about_task_adherence Before code modification Serena validation before reporting task completion think_about_whether_you_are_done Before reporting completion to user Mandatory reflection tool calls as explicit workflow steps (not just checkpoint definitions) Validate search completeness Serena think_about_collected_information Search quality validation true Validate task adherence Serena think_about_task_adherence Task alignment validation true Validate task completion Serena think_about_whether_you_are_done Completion validation true After search sequence of 3+ operations (find_symbol, search_for_pattern, get_symbols_overview, Grep, Glob) Before any symbol editing (replace_symbol_body, insert_before_symbol, insert_after_symbol, rename_symbol) or file modification Before returning final result to user Expand search scope, use alternative search strategies, or ask user for clarification Review requirements, document deviation rationale, or ask user before proceeding Identify incomplete items, iterate on missing work, or report partial completion with remaining items Tool output indicates incomplete coverage, missing critical areas, or low confidence in search results Tool output indicates task deviation, requirements mismatch, or scope creep detected Tool output indicates incomplete items, pending work, or unresolved blockers Failure-handling phase template for exceptional paths aligned with shared workflow usage Handle errors and edge cases gracefully If tool call fails: log error and attempt alternative approach If data unavailable: document gap and continue with bounded analysis If contradictory evidence: flag uncertainty and request clarification Standard agent reference syntax for commands

Use ref attribute to reference agent defined in agents/ directory. readonly attribute indicates whether agent can modify files.

Standard self-evaluation phase for commands and agents that produce reports Brief quality assessment of output Calculate confidence using decision_criteria factors Decision criteria evaluation Confidence score Identify top 1-2 critical issues if confidence below 80 Gap analysis Issue list Append self_feedback section to output Output formatting Self-feedback section Standard self-feedback output section for commands that include self-evaluation XX/100 (based on decision_criteria calculation) - [Critical] Issue description (if any, max 2 total) - [Warning] Issue description (if any)

<best_practices> Use output_format for all agents that return structured results Include reflection_checkpoint at key workflow decision points Include prepare_phase for Serena initialization in commands Add serena_validation to checkpoints before code modifications Use reflection_workflow_steps pattern for mandatory think_about_* calls Add self_evaluate_phase for agents producing reports or recommendations Use failure_handling phase in all workflows Use agent_ref syntax for consistent agent references in commands </best_practices>

Output status must use standard criteria (success >= 80, warning 60-79, error less than 60) Reflection checkpoints must include confidence threshold Commands must include prepare_phase for Serena initialization Use serena_validation_pre_edit before code modification operations Call think_about_collected_information after any search sequence of 3+ operations Call think_about_task_adherence before any symbol editing or file modification Call think_about_whether_you_are_done before returning final result to user Include failure_handling phase in complex workflows Use self_feedback_output format for self-evaluation results Use agent_ref with readonly attribute for clarity Add serena_validation_investigation after investigation phases Use standard output_status_criteria thresholds Include confidence score in all structured outputs Define threshold in reflection_checkpoints Include prepare_phase in command workflows Use serena_validation before code modifications Call think_about_collected_information after 3+ search operations Call think_about_whether_you_are_done before final response Custom status thresholds that differ from standard Omitting failure_handling in complex workflows Omitting Serena initialization in commands Skipping reflection tool calls at mandatory checkpoints

<related_skills> Base templates for error escalation, decision criteria, enforcement Parallel execution and timeout configuration Serena MCP operations (memory, symbol search, code navigation) </related_skills>