Record brain routing corrections with confidence scoring. Use when the brain orchestrator detects a wrong routing decision and needs to log the correction for future learning.
From para-obsidiannpx claudepluginhub nathanvale/side-quest-marketplace-old --plugin para-obsidianThis skill is limited to using the following tools:
Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Implements structured self-debugging workflow for AI agent failures: capture errors, diagnose patterns like loops or context overflow, apply contained recoveries, and generate introspection reports.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
Record routing corrections with confidence scoring. Called by the brain orchestrator when Nathan signals a wrong routing decision. Uses the Claude-Reflect pattern: corrections must occur 3+ times before being promoted to active learnings.
Receive correction context as args:
input="<original request>" routed_to="<wrong skill>" correct_skill="<right skill>" reason="<Nathan's words>"
Parse these 4 fields from $ARGUMENTS.
para_read({ file: "plugins/para-obsidian/skills/brain/references/learnings.md" })
Parse the Raw Correction Log table to find existing entries.
Search the Raw Correction Log for a row where Routed To matches routed_to AND Correct Skill matches correct_skill AND the Input Pattern is semantically similar to the new input.
Input Pattern to be a generalized pattern if the inputs differ slightly (e.g., "process inbox" and "handle my inbox" → "inbox processing requests").Use para_replace_section to update the "Raw Correction Log" section with the modified table:
para_replace_section({
file: "plugins/para-obsidian/skills/brain/references/learnings.md",
heading: "Raw Correction Log",
content: "<updated table with all rows>",
response_format: "json"
})
If the updated count reaches 3 or higher, also update the "Active Learnings" section by adding a human-readable routing rule:
para_replace_section({
file: "plugins/para-obsidian/skills/brain/references/learnings.md",
heading: "Active Learnings (count >= 3)",
content: "<existing rules>\n- When input matches \"<generalized pattern>\", route to `<correct_skill>` (not `<routed_to>`). Reason: <Nathan's reason>.",
response_format: "json"
})
Output a single confirmation line:
SKILL_RESULT:{"status":"ok","skill":"reflect","summary":"Recorded correction: <routed_to> → <correct_skill> (count: <N>)"}
If count just reached 3, add: "promoted": true to the JSON.