Tree and Status Verification
Verification date: October 10, 2026. Synthetic requests test the skill instructions; they do not establish accuracy against any real project’s tracker.
Structural Checks
Section titled “Structural Checks”Both skills passed the skill-creator quick_validate.py check for skill structure and frontmatter. Their bundled tree-format.md files are byte-identical. Each skill has its own reference and UI metadata, with no required companion skill or connector.
Forward Evaluation
Section titled “Forward Evaluation”A fresh evaluator received only Sam’s Tree, its reference, and two synthetic requests. It produced an initial course outline, an expanded Geometry follow-up, and a department tree. Review of those three outputs found familiar names, preserved relationships, correct path compression, three displayed levels, collapsed unrelated branches, and no invented statuses or sources.
A second fresh evaluator received only Sam’s Status, its reference, and two synthetic requests. It produced a review-phase view, a production-phase view with conflicting tracker snapshots, and an expanded Operations follow-up. Review of those three outputs found phase-relative Done, Needs Info for missing status, source-backed freshness judgments, no age-only stale flag on completed policy work, correct collapsed-parent summaries, and links outside the tree.
The evaluator noted that an unresolved failed deployment has no dedicated status in the four-label vocabulary. It used In progress with a failure qualifier. It also explained that snail markers referred to superseded tracker labels rather than the newer release records. Both choices fit the contract; these remain judgment areas to watch in real use.
Actual outputs are retained in Sam’s Tree evaluation and Sam’s Status evaluation.
Packaging and Docs Checks
Section titled “Packaging and Docs Checks”- Skills CLI 1.7.0 installed each skill separately into a clean temporary project. All four files in each tested skill package matched its source byte-for-byte; evaluation output records were added afterward.
- The strict Claude plugin and marketplace validators passed. JSON parsing confirmed collection version 0.9.0 and the Codex skills directory.
- The Astro docs build passed and generated the four new pages plus the agent-readable exports.
- The docs-spec gate passed all 11 checks with no warnings or failures.
- Scoped Biome 2.5.15 and oxlint 1.86.0 checks passed after formatting the touched configuration files. The workspace lint wrapper initially failed because its installed tools were incompatible or missing; isolated policy-eligible tools ran the same rules from the release checkout.
- Baseline built-in prek checks passed, and the staged Gitleaks scan found no secrets.
Deployment is a separate delivery step. These local checks do not claim a completed deployment.
Reproduce
Section titled “Reproduce”The synthetic requests and pass conditions live under each skill’s evals/requests.md. Give an evaluator the skill, its reference, and a request without the pass conditions. Retain the output, then assess the conditions separately.
Limits
Section titled “Limits”These are bounded synthetic smoke evaluations, not a cross-model benchmark. Live Linear, Notion, Jira, and Basecamp integration is not exercised. Freshness and phase inference remain judgments that depend on the quality of available evidence.
October 14 Visual-Marker Correction
Section titled “October 14 Visual-Marker Correction”After a reported unmarked-tree output, the skill now requires meaningful visual markers on every row. A fresh evaluator produced a category-emoji tree and a no-emoji follow-up using text markers. Both preserved the hierarchy, collapsed unrelated branches, and avoided invented status. Both skill validators passed, global instructions matched canonical instructions, and the bundled references matched. The correction was validated locally on October 14 and prepared for repository publication on October 15. Docs deployment is a separate action.