Purpose: help agents, maintainers, and contributors know which files to trust first in affiliate-skills. Scope: skill definitions, shared references, registry,
The Agent Source Of Truth skill establishes a clear hierarchy of trusted files and references within the affiliate-skills repository. It guides agents, maintainers, and contributors on which documentation and code files to prioritize when determining the definitive behavior, inputs, outputs, and constraints of individual skills. This ensures consistent interpretation across skill definitions, shared references, registries, tooling, and evaluation protocols by specifying canonical sources and rules for resolving conflicts.
By codifying trust rules, such as prioritizing `SKILL.md` files over README or QUICKSTART documents, and relying on shared references like compliance and strategy documents, this skill reduces ambiguity and drift. It also clarifies the role of machine-readable catalogs and tooling code as final authorities in their respective domains, supporting accurate skill execution and maintenance.
This skill is essential for growth leads managing complex affiliate marketing stacks where multiple contributors maintain skills with overlapping or evolving documentation. Performance marketers who rely on accurate skill definitions to automate campaign optimizations benefit from the clarity it provides about source authority. Agency strategists coordinating cross-team efforts to update or audit affiliate skills also depend on it to resolve discrepancies and maintain compliance consistently.
In scenarios where contributors submit updates or new skills, this skill helps maintainers verify changes against the canonical sources to prevent workflow disruptions or stale data. It is also useful for SEO/PPC operators integrating third-party affiliate data who need confidence in the accuracy of skill metadata and evaluation standards.
Practitioners begin by consulting the `skills/*/*/SKILL.md` file to understand the core behavior and contract of a skill, treating it as the primary source. Next, they examine skill-local references under `skills/*/*/references/*` to get detailed or supporting context. When broader doctrine or shared terminology is involved, they turn to shared references in `shared/references/*`, such as compliance or offer frameworks.
For cross-skill validation, users rely on the machine-readable `registry.json` and verify it against actual skill files to detect drift, correcting as necessary. When working on tooling or runtime behavior, they prioritize inspection of the source code in `tools/src/*`, treating public docs like README or QUICKSTART as secondary guidance. Finally, for quality assurance, they use evaluation files under `evals/` and test commands to confirm correctness.
Which file should I trust first to understand a skill’s inputs and outputs? Always start with the skill’s `SKILL.md` file. How do I resolve conflicts between public docs and runtime behavior? Trust the tooling source code in `tools/src/*` over README or QUICKSTART files. What if the registry catalog disagrees with actual skill files? Treat this as registry drift and fix it intentionally by aligning the registry to the skill files.
Attach this skill to your Metaflow agent task to enforce consistent source-of-truth rules when loading and interpreting affiliate skills. The agent will prioritize skill metadata and references based on the defined hierarchy, reducing errors from conflicting or outdated documentation. This ensures your workflows rely on the most authoritative definitions available within the repository. Use this as a foundation for...
For broader context, see our roundup of claude skills marketing, and read how to create Claude skills for related setup guidance.