drift-reconcile
Use when ConfigHub's Unit Data and the cluster's live state for that Unit have diverged — phrases like "reconcile drift", "the cluster changed out of band", "someone kubectl edit'd this", "ConfigHub and the cluster disagree", "accept the live changes", "overwrite the cluster with ConfigHub", "refresh from live", "who owns this drift?", "we have drift on app-a in prod". Runs `cub unit refresh` to pull current live state, `cub unit diff` against Data, walks the decide-who-wins decision (ConfigHub wins → re-apply; cluster wins → absorb; merge → selective reconcile), and executes the chosen path. Do not load for revision history rewind (use `rollback-revision`), for post-apply verification / three-way agreement checks (use `verify-apply`), for the first-time apply of a newly-bound Unit (use `cub-apply`), or for importing wholesale live resources into a brand-new Unit (use `import-from-cluster`).
89
88%
Does it follow best practices?
Impact
Pending
No eval scenarios have been run
Passed
No known issues
Quality
Discovery
100%Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.
This is an excellent skill description that excels across all dimensions. It provides rich, natural trigger phrases that mirror how users actually talk about drift reconciliation, lists concrete CLI commands and decision paths, and explicitly delineates its boundaries against four related skills. The description is thorough without being padded, and uses proper third-person voice throughout.
| Dimension | Reasoning | Score |
|---|---|---|
Specificity | Lists multiple specific concrete actions: runs `cub unit refresh` to pull live state, `cub unit diff` against Data, walks the decide-who-wins decision with three explicit paths (ConfigHub wins → re-apply; cluster wins → absorb; merge → selective reconcile), and executes the chosen path. | 3 / 3 |
Completeness | Clearly answers both 'what' (refreshes live state, diffs against Data, walks decide-who-wins decision, executes chosen path) and 'when' (explicit 'Use when' clause with extensive trigger phrases). Additionally includes explicit 'Do not load for' negative triggers distinguishing it from related skills. | 3 / 3 |
Trigger Term Quality | Excellent coverage of natural trigger phrases users would actually say: 'reconcile drift', 'the cluster changed out of band', 'someone kubectl edit'd this', 'ConfigHub and the cluster disagree', 'accept the live changes', 'overwrite the cluster with ConfigHub', 'refresh from live', 'who owns this drift?', and a concrete example 'we have drift on app-a in prod'. | 3 / 3 |
Distinctiveness Conflict Risk | Exceptionally distinctive — explicitly names four related skills it should NOT be confused with (`rollback-revision`, `verify-apply`, `cub-apply`, `import-from-cluster`) and provides clear negative boundaries, making conflict with other skills very unlikely. | 3 / 3 |
Total | 12 / 12 Passed |
Implementation
77%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is a high-quality, deeply actionable skill that handles a genuinely complex decision-tree workflow with clear structure and concrete commands at every step. Its main weakness is length — at ~250+ lines it pushes the boundary of what should live in a single SKILL.md, with some sections (Worker elision details, source taxonomy) that could be extracted to supporting files. The workflow design is excellent, with preflight gates, stop conditions, dry-run patterns, and a verify chain that together make the multi-step process safe and auditable.
Suggestions
Consider extracting the 'What the Worker already elides' section into a referenced file (e.g., worker-elision.md) to reduce the main skill's token footprint while preserving the information for when it's needed.
The 'Do not load for' section partially duplicates the frontmatter description — trim it to a brief cross-reference table rather than re-explaining each case.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is thorough and mostly earns its length given the complexity of the decision tree, but some sections are verbose — e.g., the 'What the Worker already elides' section explains controller behavior at length, and 'Name the likely source' enumerates categories that could be tighter. The 'Do not load for' section partially duplicates the frontmatter description. Still, most content is domain-specific knowledge Claude wouldn't have, so it's not explaining basics. | 2 / 3 |
Actionability | Excellent actionability throughout. Every resolution path has concrete, copy-paste-ready CLI commands with proper flags. The dry-run workflow, the refresh-then-restore audit trail pattern, the selective merge with function set, and the bulk commands are all fully executable. The diff/livedata/livestate diagnostic commands are specific and well-explained. | 3 / 3 |
Workflow Clarity | The five-step loop (identify → name source → decide → resolve → close recurrence) is clearly sequenced with explicit validation checkpoints. Each resolution path (4a/4b/4c) has its own sub-workflow. Preflight gates, stop conditions, dry-run-first guidance for bulk operations, and a verify chain all provide strong feedback loops. The tool boundary section prevents destructive mistakes. | 3 / 3 |
Progressive Disclosure | The skill references companion skills (cub-apply, cub-mutate, verify-apply, rollback-revision) and external references (cub-cli.md, functions-catalog.md, docs URL) appropriately. However, the content is quite long and monolithic — the Worker elision details, the source-naming taxonomy, and the bulk operations section could potentially be split into referenced files. Without bundle files provided, the references to 'references/cub-cli.md' and 'references/functions-catalog.md' can't be verified. | 2 / 3 |
Total | 10 / 12 Passed |
Validation
81%Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.
Validation — 9 / 11 Passed
Validation for skill structure
| Criteria | Description | Result |
|---|---|---|
allowed_tools_field | 'allowed-tools' contains unusual tool name(s) | Warning |
frontmatter_unknown_keys | Unknown frontmatter key(s) found; consider removing or moving to metadata | Warning |
Total | 9 / 11 Passed | |
- Repository
- confighub/confighub-skills
- Commit
59ea831
- Reviewed
Table of Contents
Is this your skill?
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.