cortex-kanban-worker

name	cortex-kanban-worker
description	Use when a Kanban worker has Cortex tools available and needs to retrieve, apply, and cite Vault knowledge while executing a task. Covers vault_search, vault_read_note, and vault_build_context workflows for task context, research, and completion summaries.
version	1.0.0
tags	["kanban","cortex","vault","research","worker"]

Cortex Kanban Worker — Vault-backed task execution

This skill is loaded for Hermes Kanban workers that run with the Cortex plugin and Vault tools available (HERMES_KANBAN_TASK is set).

Overview

As a Kanban worker with Cortex tools, use the Vault as durable project memory while executing a task: retrieve relevant knowledge, inspect source notes, apply the findings to the work, and record which Vault evidence was used.

Workflow

1. Read the task and load context

# First step: inspect the current Kanban task.
task = kanban_show()
title = task["title"]
body = task.get("body", "")

# Load relevant Vault context for the task topic.
context = vault_build_context(
    query=f"{title} {body}",
    budget=1500,        # roughly 1500 tokens of context
)
# context.text contains ranked Vault excerpts.

2. Run targeted Vault research

# Search for specific information.
results = vault_search(
    query="search terms derived from the task context",
    top_k=10,
    filters={"type": ["fact", "decision"]},  # optional: restrict note types
)
for r in results["results"]:
    note = vault_read_note(file=r["file"])
    print(note["content"])  # full note text

3. Record findings in the task result

kanban_complete(
    summary="Research completed — 3 relevant Vault notes found and evaluated",
    metadata={
        "vault_notes_consulted": [
            "10_facts/example.md",
            "30_projects/example.md",
        ],
        "vault_search_queries": ["search terms used"],
        "findings": "short summary of the findings",
    },
)

Patterns

Pattern A: Research task

Task: "Research X and summarize the findings"
  → vault_build_context(query="X", budget=2000)
  → vault_search(query="X", top_k=5)
  → For each relevant note: vault_read_note()
  → Return findings via kanban_complete(metadata={...})

Pattern B: Add Vault context to an implementation task

Task: "Build feature Y"
  → vault_build_context(query="Feature Y architecture", budget=2000)
  → vault_search(query="Feature Y similar projects")
  → vault_read_note() for relevant architecture decisions
  → Use the retrieved knowledge while implementing the feature

Pattern C: Decision support from the Vault

Task: "Choose between option A and option B"
  → vault_search(query="Option A comparison", filters={"type": ["decision"]})
  → vault_search(query="Option B lessons learned")
  → vault_read_note() for matching decisions
  → Include the evidence in the task rationale

Filters for vault_search / vault_build_context

Filter	Values	Effect
type	fact, decision, project, runbook, map, task	Restrict results to selected note types
tags_any	["cortex", "hermes"]	Notes with at least one listed tag
tags_all	["cortex", "kanban"]	Notes with all listed tags
domain	"development", "infrastructure"	Notes in one domain
project	"hermes-cortex"	Notes for one project
importance_min	3	Notes with importance >= 3

Notes

• Cache warming may run at session start, but the first vault_search call can still be slower than later calls.
• vault_build_context is the most efficient overview path: one call, ranked and token-budgeted.
• Use vault_search + vault_read_note for deep dives into individual notes.
• Results include scores and ranks; Cortex ranking is hybrid (BM25 + vector + graph).
• Record the Vault notes and search queries used in kanban_complete(metadata=...) so reviewers can trace conclusions back to evidence.