Status: Accepted
Date: 2026-07-06

Context

.context/findings/index-yaml-split-review-2026-07-06.md’s Migration-Risk review found .tessl/plugins/** — the vendored, installed mirror of every helper skill this repo authors — is entirely gitignored (.gitignore line 11), and therefore structurally invisible to any git-based CI check. That review’s own worked example (context-index’s mirror) had already diverged once. Run through design-debate (Advocate / Skeptic / Migration-Risk) to decide what, if anything, to do about it.

Decision

1. Do not track .tessl/plugins/pantheon-org/** in git. Tracking was seriously considered (436KB/74 files, small, with real precedent in this repo for tracking generated content — .context/audits/) but Migration-Risk’s review found tracking alone is inert: nothing verifies the tracked copy matches a fresh tessl install without also writing a new CI step. Once that CI step is required regardless, tracking adds cost (a fragile .gitignore negation pattern — a naive one does nothing; the correct layered form is non-obvious — plus a doubled diff on every skill-content PR) without adding protection beyond what the CI step alone provides.
2. Instead, verify via a CI-only ephemeral diff: a new step runs tessl install into a scratch location and diffs the result against .context/plugins/pantheon-org/**'s expected install output, failing the build on divergence. No content is committed; no .gitignore change is needed.
3. Scope is pantheon-org/** only. Third-party registry content (.tessl/plugins/pantheon-ai/** at ~174MB, .tessl/plugins/tessl-labs/**) is explicitly out of scope for both the rejected tracking option and the accepted diff-check option — we don’t author it, and its size alone would make tracking a non-starter if it were ever proposed.
4. Not yet implemented. This ADR records the design decision (which approach) reached via design-debate; the CI step itself is a follow-up, natural input to plan-create.

Consequences

• Easier: drift between an authored pantheon-org skill and its installed mirror becomes CI-catchable without any repo bloat, .gitignore fragility, or doubled review surface.
• Harder: until the CI step is actually built, the underlying gap (mirror drift is invisible) remains open — this decision picks the mechanism, it doesn’t close the gap yet.
• Binding for future work: if someone later proposes tracking .tessl/plugins/** (any subset) to solve a drift concern, this ADR’s reasoning — tracking is inert without a verification step, and the verification step alone is sufficient — should be addressed directly, not re-litigated from scratch.