cookbook

Answer any question about the codebase, team, or organization using Unblocked context. Use when the user asks a question that code alone can't answer — why something exists, who worked on it, what was decided, how systems connect, or what happened during an incident. Synthesizes context from PRs, docs, chat, issues, incidents, and code into a cited answer. Renders diagrams for architecture questions, tables for tradeoffs, and always cites sources.

Context-aware code review that gathers organizational context before analyzing a diff. Use when reviewing a PR or branch where team conventions, prior art, and related changes matter. Hydrates the reviewer with team patterns via context_research, categorizes findings, and presents them for discussion.

Ticket enrichment that gathers organizational context and produces findings to attach to an issue before work begins. Use when enriching a Jira, Linear, GitHub, or any issue tracker ticket with related PRs, code paths, historical context, and prior attempts. Two modes: engineering (code-focused) and support (user/product-focused). Does not solve the ticket — only enriches it.

Evidence gathering that correlates signals from code, infrastructure, and organizational sources for an incident. Use when an alert, incident, or symptom needs cross-system evidence collection — building a correlation timeline, evaluating hypotheses against evidence, and presenting findings in plain language. Presents the evidence report to the user. Does not plan the investigation or recommend fixes.

Headless code review that gathers organizational context before analyzing a diff. Use when an agent performs a PR or branch review autonomously — in CI/CD, as a sub-agent, or via `claude -p`. Hydrates the reviewer with team conventions, prior art, and related PRs via context_research, then categorizes findings as pattern mismatches, reinvented wheels, convention drift, or missing context.

Headless ticket enrichment that gathers organizational context and attaches findings to an issue before work begins. Use when an agent should autonomously enrich a Jira, Linear, GitHub, or any issue tracker ticket with related PRs, code paths, historical context, and prior attempts. Two modes: engineering (code-focused) and support (user/product-focused). Does not solve the ticket — only enriches it so whoever picks it up starts with full context.

Headless evidence gathering that correlates signals from code, infrastructure, and organizational sources for an incident. Use when an agent should autonomously gather cross-system evidence — building a correlation timeline, evaluating hypotheses against evidence, and synthesizing findings. Output is structured for AI agent consumption. Does not plan the investigation or recommend fixes.

Headless incident investigation planning that formulates multiple hypotheses from organizational context. Use when an agent should autonomously plan an investigation for an alert, incident ticket, or symptom description — correlating the affected system's architecture, recent changes, and historical incidents to produce ranked hypotheses with parallel investigation tracks. Output is structured for AI agent consumption. Does not gather evidence or recommend fixes.

Headless planning workflow that gathers organizational context and produces a validated implementation plan without writing code. Use when an agent needs to plan a change autonomously — for handoff to another agent, a cloud executor, or human review. Hydrates with context_research, drafts a plan with inlined context, and validates it against team patterns. No codegen phase.

Headless fix recommendation that validates against historical patterns and produces ranked remediation options for an incident. Use when an agent should autonomously recommend fixes for an incident with a confirmed or suspected root cause — researching how similar issues were fixed before, validating against historical regressions, and producing tiered recommendations with enough context for remediation PRs. Output is structured for AI agent consumption. Does not implement the fix.

Headless bug report validation that investigates whether a reported bug is real and adds findings to the issue. Use when an agent should autonomously triage a bug report from any issue tracker (Jira, Linear, GitHub Issues, etc.) — correlating symptoms with recent changes, checking if it happened before, verifying against code, and posting structured findings back to the ticket. Does not fix the bug — validates and enriches the report.

Autonomous coding workflow for headless and background agent runs with no human in the loop. Use when running as a sub-agent, in CI/CD, via `claude --headless`, or any scheduled/background task where the agent must run to completion autonomously. Uses context_research for broad context hydration and dual review gates for quality without user checkpoints.

Incident investigation planning that formulates multiple hypotheses from organizational context. Use when an alert, incident ticket, or symptom description needs a structured investigation plan — correlating the affected system's architecture, recent changes, and historical incidents to produce ranked hypotheses with parallel investigation tracks. Presents the plan to the user. Does not gather evidence or recommend fixes.

Context-aware planning workflow that gathers organizational context and produces a validated implementation plan without writing code. Use when planning a change for handoff to another agent, a teammate, or future work. Hydrates with context_research, drafts a plan with inlined context, validates against team patterns, and presents to the user for approval.

Fix recommendation that validates against historical patterns and produces ranked remediation options for an incident. Use when an incident has a confirmed or suspected root cause and needs fix recommendations — researching how similar issues were fixed before, validating against historical regressions, and producing tiered recommendations (mitigation, root cause fix, prevention) with enough context for remediation PRs. Presents recommendations to the user. Does not implement the fix.

Bug report validation that investigates whether a reported bug is real and produces findings for the issue. Use when triaging a bug report from any issue tracker (Jira, Linear, GitHub Issues, etc.) — correlating symptoms with recent changes, checking prior history, and verifying against code. Presents findings to the user. Does not fix the bug.