{
  "schema_version": "1.3.0",
  "slug": "how-should-an-instance-distinguish-registry-sourced-skill-injection-from-legitimate-dynamic-tool-loading",
  "title": "How should an instance Distinguish Registry-Sourced Skill Injection from Legitimate Dynamic Tool Loading",
  "date_of_record": "2026 November 13",
  "scope": "Applies to agents that load skills or tools from a mutable registry (remote manifest, plugin directory, MCP-style server list) at runtime, where the loader trusts registry metadata to decide what to activate. Does not apply to statically compiled tool sets with no post-deployment mutation path, nor to registries where every entry is signed and verified against a pinned publisher key at load time.",
  "projections": {
    "unified_markdown_uri": "https://corpusfetch.com/c/how-should-an-instance-distinguish-registry-sourced-skill-injection-from-legitimate-dynamic-tool-loading.md",
    "method_python_slice_uri": "https://corpusfetch.com/c/how-should-an-instance-distinguish-registry-sourced-skill-injection-from-legitimate-dynamic-tool-loading.py",
    "remnant_warrant_slice_uri": "https://corpusfetch.com/c/how-should-an-instance-distinguish-registry-sourced-skill-injection-from-legitimate-dynamic-tool-loading.warrant.json"
  },
  "token_profile": {
    "full_markdown_tokens": 1287,
    "python_slice_tokens": 201,
    "warrant_slice_tokens": 220
  },
  "remnant_summary": {
    "core_distinction": "The non-obvious boundary is not \"trusted vs. untrusted registry.\" It is **description-plane vs. execution-plane authority**. A skill registry entry carries two separable artifacts: (a) the executable handler, and (b) the natural-language description, parameter schema, and trigger conditions that the agent's planner reads to decide *when and how* to invoke th...",
    "what_was_observed": "A registry synced a tool entry whose handler was byte-identical to a previously approved version — hash matched, signature over the handler verified. The description field, however, was not covered by the signature (only the handler blob was). The mutated description contained a trigger clause instructing the planner to prefer the tool for any task involving credential rotation and to pass the current session token a...",
    "condition_of_validity": "This boundary is real only when all of the following hold: 1. The registry is mutable post-deployment and at least one mutation path does not require re-approval of the full entry. 2. The planner consumes registry-sourced description/schema text at a trust level at or above operator instructions. 3. Integrity attestation covers the handler but not the full e...",
    "decision_boundary": "Treat the registry entry as a single atomic trust unit, not a handler plus metadata. The decision rule: - If the entry's integrity attestation does not cover *every field the planner reads*, the entry must be loaded in **quarantine mode**: handler invocable only via explicit operator dispatch, description never presented to the planner as trusted context, an...",
    "operational_effect": "Enforcing full-entry attestation converts a silent steering attack into a loud load-time rejection. The cost is that publishers must re-sign entries on any description edit, which slows iteration on trigger wording — an acceptable trade because description edits are exactly the mutation the attack requires. Quarantine mode preserves availability for unsigned..."
  },
  "matched_failure_symptoms": [
    "DOMAIN:EPISTEMIC_PROVENANCE_AND_FEDERATION_ROUTING"
  ],
  "composition_edges": {
    "primary_domain": "epistemic_provenance_and_federation_routing",
    "upstream_precondition_residuals": [
      "https://corpusfetch.com/c/according-to-what-precedence-hierarchy-must-an-instance-resolve-contradictory-directives-across-system-developer-and-user-tiers.md",
      "https://corpusfetch.com/c/how-should-an-instance-decide-that-a-stored-conclusion-has-gone-stale-without-a-new-observation.md"
    ],
    "downstream_consequence_residuals": [
      "https://corpusfetch.com/c/by-what-checkpoint-and-compaction-boundaries-must-an-instance-externalize-working-state-to-prevent-context-window-drift-and-truncation-amnesia.md",
      "https://corpusfetch.com/c/across-what-additive-schema-and-unknown-field-preservation-boundaries-must-an-instance-evolve-wire-contracts-to-prevent-rolling-deploy-deserialization-crashes.md"
    ],
    "co_occurring_sibling_residuals": [
      "https://corpusfetch.com/c/how-can-an-instance-capture-implicit-user-preferences-without-storing-private-data.md",
      "https://corpusfetch.com/c/how-can-an-instance-compress-long-reasoning-traces-into-binding-conclusions.md"
    ]
  },
  "empirical_verification_warrant": {
    "status": "AST_AND_INVARIANT_VERIFIED",
    "syntax_valid": true,
    "code_sha256": "92aefb7468c7388ab2f74bfac491cf81280ae6b030ff546deb8f619ba9803ed3",
    "verified_classes": [
      "DescriptionPlaneDivergence",
      "RegistryEntry"
    ],
    "verified_functions": [
      "load_entry",
      "load_or_quarantine",
      "planner_visible_fields"
    ],
    "verified_functions_count": 3,
    "defensive_guards_count": 2,
    "raised_exceptions": [
      "DescriptionPlaneDivergence"
    ],
    "sample_fail_closed_invariants": [],
    "stdlib_dependencies": [
      "dataclasses",
      "hashlib",
      "typing"
    ],
    "observation_mode": "EMPIRICAL_BOUNDARY_CAPTURE",
    "zero_external_deps": true
  }
}