0001. Centralized Agent Skill Citation Enforcement
- Status: Accepted
- Date: 2026-07-28
- Deciders: Levi (User), AI Agent
Context and Problem Statement
When AI agents use a Bioconductor skill from this repository, their responses often do not make it obvious that a specific skill was executed, which skill it was, or what version was used. This lack of transparency makes it difficult for users to track which skills were responsible for a given output and complicates debugging when skills are updated. We needed a way to guarantee that agents always cite the Bioconductor skill being used (name, version, and author).
Decision
We will enforce a standard citation block in all agent responses using a centralized, single-source-of-truth approach.
Specifically, we decided to:
- Update AGENTS.md: Add a mandatory requirement for agents to include a standard citation block (
> 🛠️ **Bioconductor Skill Executed**: <name> | **Version**: <version> | **Author**: <author>). - Make
authorRequired: Change theauthorYAML frontmatter field from optional to REQUIRED inAGENTS.mdand thevalidate_skills.pyCI script to guarantee there is always an author to cite. - Update Platform Instructions: Add explicit rules to
instructions/claude.md,instructions/copilot.md, andinstructions/antigravity.mdto bootstrap agents with the citation requirement. - Avoid Duplicating Content in Skills: Instead of hardcoding the citation format prose into the
Output Formatsection of every individualSKILL.mdfile, the rule is defined globally and agents are instructed to extract the canonical values directly from each skill’s YAML frontmatter.
Alternatives Considered
- Embedding citation instructions in every
SKILL.mdfile: We considered adding a requiredOutput Formatsection to every skill file containing the exact markdown for the citation. This was rejected because it violates the repository’s principle of avoiding content duplication. Thename,version, andauthoralready live canonically in the YAML frontmatter. - Using CI to enforce the exact citation markdown string inside skills: We considered adding a python script check to look for the exact citation block inside
SKILL.mdfiles. This was rejected because checking for exact strings (with emojis) in free-form prose is brittle and mixes presentation concerns with structural validation.
Consequences
- Easier Maintenance: If the standard citation format needs to change in the future, it only needs to be updated in the platform instruction files and
AGENTS.md, not across all 13+ skill files. - Improved Transparency: Users will now clearly see when a Bioconductor skill was utilized, including exactly which version was run.
- Immediate Compliance:
authoris now a mandatory frontmatter field for all new and existing skills.