smithery.ai

drift-detector

A skill to detect and report structural, configuration, logic, and epistemic discrepancies against defining baselines.

First seen Mar 20, 2026

Installation

$ npx skills add https://smithery.ai

Similar popular skills

Related neighbors and high-traction skills in the same topics — useful to compare before installing.

Also in this package

Other skills from smithery.ai · top by installs.

npx skills add https://smithery.ai

Browse all from smithery.ai

More details

Agent compatibility

Declared targets from SKILL.md / docs. Unmarked agents are not listed — the skill may still install via the CLI.

Claude Code Not declared
Cursor Not declared
Codex Not declared
GitHub Copilot Not declared
Windsurf Not declared
Gemini CLI Not declared
Cline Not declared
OpenCode Not declared

Package contents

Files included with this skill beyond the listing page.

  • skill md SKILL.md 2,622 B
  • docs SUMMARY.md 140 B

History

  1. First seen on skills.sh
  2. First recorded snapshot · 1 installs

SKILL.md

Drift Detector Skill

Status: Operational Skill Category: Validation (Structural)

1. Skill Purpose

The drift-detector ensures the system's actual state matches its intended state as defined by baselines and epistemic contracts. It answers:

  1. "Has the system structure or config changed since last approval?" (Syntactic)
  2. "Is the system violating its own epistemic laws?" (Semantic)

2. Invocation Contract

Standard Grammar

Invoke drift-detector
Mode: <DRY_RUN | REAL_RUN | VERIFY>
Target: <path/to/check>
ExecutionScope:
  mode: <all | selectors>
  [rule_id: <OD-1 | CD-1 | ...>]
Options:
  validators: <enabled | disabled>
  config: <check | ignore>
  structure: <check | ignore>

3. Supported Modes & Selectors

A. Execution Modes

  • DRY_RUN: Identical to VERIFY, but does not trigger any potential side-effect hooks (though drift-detector is primarily read-only).
  • REAL_RUN: Execute full scan and generate a persistent [Drift Report].
  • VERIFY: Perform a read-only comparison against baselines and return a status code (PASS/FAIL).

B. Selectors

  • all: Run all 13 epistemic rules and all structural/config checks.
  • rule_id: Target a specific epistemic rule (e.g., PD-1 for permission bypass).

4. Hook & Skill Chaining

  • Chained From: Invoked as a Pre-Execution Gate by design-build-harness.
  • Chained To: If CRITICAL drift is detected, may inform monitor-trigger.

5. Metadata & State

  • Inputs: Epistemic documents, .env, project root structure, source code.
  • Outputs: JSON Drift Report, violation count.

6. Invariants & Prohibitions

  1. Read-Only: NEVER modifies files; only reports drift.
  2. No Interpretation: Judges against baselines, not "intent" or "quality."
  3. Deterministic: Two runs on the same state MUST produce the same report.

7. Example Invocation

Invoke drift-detector
Mode: VERIFY
Target: c:\GIT\TraderFund\
ExecutionScope:
  mode: all
Options:
  validators: enabled
  config: check
  structure: check

8. Epistemic Validation Rules

Category Rules What They Detect
Drift Taxonomy OD-1 to LD-1 Ontological, causal, boundary, permission, temporal, latent→active drift
Macro/Factor MF-1 to MF-4 Macro dependency, factor policy, execution inference
Harness Safety EH-1 to EH-3 Belief inference, factor bypass, implicit strategy