Relationships
#1104 Add a 'verify capability exists before relying on it' rule to repo-init CLAUDE.md
Opened by jeremy · 7/12/2026
Problem
Skill/prompt documentation referenced workflows (via
swamp workflow run <name>) that were never actually built or
published. Each was a silent failure -- not discovered until an agent
actually tried to run the workflow, well after the documentation
referencing it had already been trusted and acted on (and, in one
case, propagated into further downstream instructions).
Nothing in the current swamp-repo-init-scaffolded CLAUDE.md/AGENTS.md rule set tells an agent to verify a referenced capability actually exists before relying on it or documenting it further. The rule set already covers several related operational-discipline habits (verify before destructive operations, use swamp rather than bypass it) but not this one.
Proposed solution
Add a new rule to the repo-init-scaffolded template: "Verify documented
capabilities before relying on them. If a skill, prompt, or instruction
references swamp workflow run <name> or a specific model method,
confirm it actually exists (swamp workflow search, swamp model type describe) before treating it as available -- documentation drifts from
what was actually built and published faster than code does."
Alternatives considered
Relying on workflow validate/workflow run to fail loudly at
execution time -- rejected because that only helps once someone
actually tries to run it. The goal is preventing an agent from
confidently trusting -- or re-documenting -- a capability that was
never real in the first place, before that trust propagates further.
Closed
No activity in this phase yet.