debug-issue

name	debug-issue
description	Systematically debug issues using graph-powered code navigation

Debug Issue

Use the knowledge graph to systematically trace and debug issues.

Installed Search Mode

This packaged skill is mode-neutral. dagayn install rewrites this section with the selected embedding mode so bug searches balance semantic recall with speed.

Steps

1. Run get_minimal_context_tool(task="<bug or symptom>") to check graph freshness, risk, and suggested next tools.
2. Use semantic_search_nodes_tool to find code related to the issue.
3. Use query_graph_tool with callers_of and callees_of to trace call chains.
4. Use traverse_graph_tool only after selecting a concrete suspected node and only when a bounded neighborhood is more useful than a specific caller, callee, import, test, or documentation relationship.
5. Use flow_tool(mode="list", detail_level="minimal") or review_tool(mode="affected_flows") to identify candidate flow names before calling flow_tool(mode="get").
6. If the suspected code point has documentation bridges, use query_graph_tool(pattern="docs_for", target="<path::symbol>", detail_level="minimal") to pull in linked specs, runbooks, explanations, and issue notes. If the bug report starts from a Markdown contract section, use query_graph_tool(pattern="implementations_of", target="<doc.md>::<section-slug>", detail_level="minimal") to find the implementation nodes linked by implemented_by / implements_contract.
7. Run review_tool(mode="changes") to check if recent changes caused the issue. Read analysis_summary for risk reasons, affected-flow rankings, hotspot proximity, and recommended tests.
8. Use review_tool(mode="impact") on suspected files only when analysis_summary or the call trace leaves the blast radius unclear.
9. Read source directly once graph evidence identifies the likely failing path.

Tips

• Check both callers and callees to understand the full context.
• Prefer relationship queries over raw traversal when the question names a relationship. Raw traversal is for "what else is nearby?" after the likely node is known.
• Look at affected flows to find the entry point that triggers the bug.
• Recent changes are the most common source of new issues.
• When the symptom starts from a log line, CLI command, or UI action, use rg once to map that literal string to a graph node, then return to graph tools.
• Treat dagayn: documentation directives as typed traceability evidence. Code comments such as # dagayn: explained-by docs/runbook.md#Failure Mode can point to runbooks or problem statements that are more useful than another caller hop.
• For documentation bridge query results, check evidence_type: authored contract links are stronger than extracted explanatory links, while heuristic_reachable links require source confirmation.
• If query_graph_tool returns no result or status="not_found", read zero_result_reason, next_action, answerability, and missingness before ruling out a path.
• Do not infer root cause from graph centrality alone; require an observed failing path, changed behavior, or source-level defect.

CLI Fallback

Use MCP tools first. If the current MCP server profile does not expose a tool such as flow_tool or review_tool, run the same implementation through the CLI without restarting the agent:

dagayn tool get_minimal_context_tool --arg 'task="debug login timeout"'
dagayn tool flow_tool --arg mode='"list"' --arg detail_level='"minimal"'
dagayn tool flow_tool --arg mode='"get"' --arg 'flow_name="handle_request"'
dagayn tool review_tool --arg mode='"impact"' --arg 'changed_files=["src/auth.py"]'
dagayn tool query_graph_tool --arg pattern='"docs_for"' --arg target='"src/auth.py::handler"'
dagayn tool query_graph_tool --arg pattern='"implementations_of"' --arg target='"docs/auth.md::login-contract"'

Token Efficiency Rules

• ALWAYS start with get_minimal_context_tool(task="<your task>") before any other graph tool.
• Use detail_level="minimal" on all calls. Only escalate to "standard" when minimal is insufficient.
• Target: complete any review/debug/refactor task in ≤5 tool calls and ≤800 total output tokens.