SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsmage-hand

mage-hand

Office & productivity

View repository

Updated 10-20 16:16

About this skill

Mage Hand

Manipulate files, records, and lightweight system state with precision and minimal blast radius.

Sigil

     __
  .-'  `-.
 /  .--.  \\
|  / /\\ \\ |
|  | \\/ | |
 \\  `--'  /
  `-.__.-'

What This Skill Does

Use this skill for small, careful remote manipulations where dexterity matters more than force. In this grimoire, Mage Hand is treated as a hybrid spell with a shipping-now delivery profile. Canonical reference input: Mage Hand (spell).

When To Use

  • Trigger this spell when the request asks for surgical, narrow edits to existing objects with explicit boundaries on what NOT to touch. Look for:
  • Specific identifiers: file paths, ticket IDs, record keys, field names, audience names
  • Constraint language: "only", "but don't touch", "leave X alone", "if X then stop"
  • Single-field or single-value changes: "change max_attempts from 3 to 4", "patch reply_to_email"
  • Conditional execution with early-exit: "only if each has approval, otherwise stop and return the exception list"
  • Audit expectations: "show the diff", "tell me the exact line", "report skipped IDs"
  • Small batch operations with per-item validation: "move tickets A, B, C but skip any missing X"

Prerequisites

  • No extra runtime dependencies beyond Hermes Agent and the normal toolset for this session.

Procedure

  1. Restate the target, the success condition, and any no-touch boundaries before taking action.
  2. Restate boundaries before acting: Name the exact target, the success condition, and what must NOT be touched.
  3. Scope to the smallest edit surface: Change one field, one line, one ticket state. Do not rewrite surrounding content.
  4. Apply with diff or state proof: Show what changed. Confirm adjacent values are untouched.
  5. Report exceptions, not just successes: If any item in a batch fails validation or lacks a prerequisite, stop and return the failure list. Do not silently skip.
  6. Package the result as the deliverables below, with confidence, assumptions, and unresolved risk called out explicitly.

Deliverables

  • A minimal, well-scoped change.
  • A short audit trail of touched surfaces.
  • A note on adjacent objects that were intentionally left alone.

Pitfalls / Guardrails

  • Keep the theatrical framing, but name the concrete mechanism that makes the skill useful right now.
  • Do NOT route here when the request involves:
  • Do not use for: Building new interfaces or features ("give this script a chat front end", "create a dashboard")
  • Do not use for: Setting up recurring automation ("every Friday", "whenever X happens, do Y automatically")
  • Do not use for: Debugging, forensics, or tracing ("find what changed", "trace which integration sent this")
  • Do not use for: System recovery or access restoration ("token expired, find the recovery path")
  • Do not use for: Performance tuning or capacity changes ("slow down the queue", "scale up workers")
  • Do not use for: Broad refactors or rewrites ("rewrite this module", "restructure the config")
  • Do not use for: The key differentiator: Mage Hand moves or edits existing specific objects with known targets. It does not build, automate recurring tasks, investigate, or recover.

Verification

  • Check that the result includes every deliverable promised above.
  • Check that confirmed facts, assumptions, and inferences are visibly separated.
  • Check which parts are concrete actions versus framing, so the user can tell what is real now.

Example Invocation

/mage-hand make the smallest safe change needed here and show exactly what you touched