Content
77%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured project-policy skill: efficient tables, concrete values, and workflows with genuine error-recovery loops and pre-conditions. Its main defect is bundle integrity — the three tool reference files it repeatedly points to (for all create/update/transition syntax) are missing from the bundle, which undermines both navigation and executability of the deferred operations.
Suggestions
Ship the three referenced files (references/cli.md, references/mcp-atlassian.md, references/devprod-gateway.md) in the skill bundle — the body delegates all per-tool syntax to them, so without them no create, transition, or label operation is executable.
Collapse the three-row tool-preference memory table into a single naming pattern with one concrete example (e.g., the CLI row) to cut roughly a third of that section's tokens without losing information.
Inline at least one minimal executable example per backend (e.g., a single jira CLI create command) so the skill degrades gracefully if a reference file is unavailable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and table-driven, carrying only project-specific policy Claude cannot infer (labels, statuses, story-point scale, link types, custom field IDs). Minor trimming is possible — the tool-preference memory table spells out three full rows (filename, slug, body, MEMORY.md line) where a naming pattern plus one example would convey the same information. | 4 / 5 |
Actionability | Concrete and specific throughout: probe command "which jira && jira me" with a 5-second timeout, exact field "customfield_12054", enumerated label/status/resolution values, and a copy-paste follow-up description block. The gap is that all per-tool execution syntax is deferred to the tool reference files, so the body alone cannot execute a create, transition, or label-set. | 4 / 5 |
Workflow Clarity | Multi-step processes are explicitly sequenced with validation checkpoints and feedback loops: the numbered tool-selection flow with probe order and memory persistence, the quick-fail rule ("Do not retry the failed tool more than once... If none succeed, report the error"), pre-close checks (check Story Points before attempting the transition rather than reacting to failure; confirm the PR merged), and confirmation gates before any destructive ticket modification. | 5 / 5 |
Progressive Disclosure | The delegation pattern is right — per-tool syntax is clearly signaled and one level deep via [references/cli.md], [references/mcp-atlassian.md], and [references/devprod-gateway.md], keeping the overview lean. However, the bundle contains no references/ directory, so every referenced path fails to resolve and the navigation the skill depends on does not exist; scoring against the actual bundle structure, this falls to the mid anchor. | 3 / 5 |
Total | 16 / 20 Passed |