MCPTools
piyaz_context
Retrieve task context at varying depth.
piyaz_context
Retrieve task context at varying depth. ALWAYS fetch context before reasoning about a task; pick the lightest depth that answers the question. summary=task header + description + counts (criteria, decisions, plan flag, edge counts) + full 1-hop edges WITH notes. The lightest depth that still carries edge notes; folds in what piyaz_query type='edges' would give. working=detailed (criteria, decisions, 1-hop edges) for refinement and review. agent=multi-hop dependency chains with upstream execution records (~4-8K tokens); fetch BEFORE coding. planning=spec-focused (project description, prereqs, acceptance criteria, downstream specs); fetch BEFORE writing the implementation plan.
Actions
| Action | Purpose |
|---|---|
summary | task header + description + counts + 1-hop edges with notes (folds in piyaz_query type='edges') |
working | criteria, decisions, 1-hop edges (both depends_on and relates_to, both directions, with notes), does NOT render executionRecord, files, or implementationPlan |
agent | multi-hop deps + upstream execution records + files + downstream; renders the task's own executionRecord when status is done/cancelled (use BEFORE coding, and to read a finished task's record) |
planning | project description, prereqs, ACs, downstream specs (use BEFORE writing the implementation plan) |
review | in_review review bundle: implementationPlan alongside executionRecord, PR link surfaced, plan-vs-files drift, AC evaluation, downstream impact, review-lens prompts (security / perf / reliability / observability / codebase standards). The review subagent reads this depth |
Parameters
| Name | Type | Required | Description |
|---|---|---|---|
depth | "summary" | "working" | "agent" | "planning" | "review" | No | summary=task header + description + counts + 1-hop edges with notes (folds in piyaz_query type='edges'). working=criteria, decisions, 1-hop edges (both depends_on and relates_to, both directions, with notes), does NOT render executionRecord, files, or implementationPlan. agent=multi-hop deps + upstream execution records + files + downstream; renders the task's own executionRecord when status is done/cancelled (use BEFORE coding, and to read a finished task's record). planning=project description, prereqs, ACs, downstream specs (use BEFORE writing the implementation plan). review=in_review review bundle: implementationPlan alongside executionRecord, PR link surfaced, plan-vs-files drift, AC evaluation, downstream impact, review-lens prompts (security / perf / reliability / observability / codebase standards). The review subagent reads this depth. |
projectId | string (uuid) | No | Project UUID. Required for 'working' depth. |
taskId | string (uuid) | Yes | Task UUID. |