Validate Athena research commits for evidence and judgment layer separation.
npx skills add https://github.com/harrymcdonagh/athena --skill wall-guard
Please help me install this Agent Skill. Skill: wall-guard Source: https://github.com/harrymcdonagh/athena/tree/main/.claude/skills/wall-guard Command: npx skills add https://github.com/harrymcdonagh/athena --skill wall-guard
This Skill ensures that research commits in Athena adhere to strict evidence and judgment separation rules, maintaining the integrity of research outputs.
Run the wall-guard skill before any commit touching research endpoints to enforce the evidence/judgment wall.
High-intent search queries and answers about installing and using this skill.
To enforce evidence and judgment separation in research commits, validate commits affecting research endpoints, answer schemas, prompts, or retrieval code to block any ranking, scoring, or evaluative language. This maintains strict research output integrity.
An evidence layer in research codebase validation is a structured data schema that must remain free of verdict fields and subjective judgment. It requires presenting findings, such as FIND results, strictly ordered by match strength.
To check commits for evaluative language before submission, run a pre-commit validation against research endpoints, answer schemas, prompts, and retrieval code to ensure no ranking or scoring terms are present in the evidence layer.
Yes, separating the judgment layer requires its own ADR and must use only read-only market data. This ensures subjective evaluation logic remains structurally isolated from the evidence layer in the research codebase.
Commit validation rules are triggered by changes affecting research endpoints, answer schemas, prompts, or retrieval code. These specific areas must be checked to maintain the strict evidence and judgment separation wall.
Verdict fields are blocked in evidence-layer schemas to prevent evaluative judgment from contaminating objective research data. Enforcing this separation ensures that FIND results remain ordered by match strength without subjective scoring interference.
FunnelEnvyQuery Google Gemini CLI for code analysis and codebase exploration.
jabrenaGenerate MADR-compliant Architecture Decision Records for Java projects.
JackSmack1971Configure pre-commit hooks to validate conventional commit messages.
CsHengEnforce LLM content governance and compliance across LLM-facing files.
