Content
70%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.
An operationally strong skill body: every sub-command has concrete commands, explicit validation checkpoints, and confirmation gates for shared-resource writes, and the conflict-resolution and mutual-exclusion rules show real domain judgment. The main weaknesses are redundant token-security material repeated across four sections, working-directory inconsistencies in the push/status command blocks, and a monolithic 210-line file that neither bundles nor splits out the scripts and detail it depends on.
Suggestions
Consolidate the token-security material: the setup section's 5-step list and the 'Token Security' defense-in-depth table restate the same guards — merge them into one presentation (or move the table to references/security.md) to cut roughly 30 duplicated lines.
Fix the working-directory inconsistency in the push and status blocks: after 'cd paper-overleaf && git pull --ff-only', the sibling-relative paths ('rsync ... paper/ paper-overleaf/', 'diff -rq --brief paper/ paper-overleaf/') no longer resolve — anchor all commands to the repo root or use explicit parent-relative paths.
Split detail-heavy guidance (the per-hunk diff protocol and conflict-resolution rules) into a references/ file so SKILL.md stays an operational overview, and bundle overleaf_setup.sh / overleaf_audit.sh in scripts/ so the security guards the skill relies on actually ship with it.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly tight and operational (no explanations of git/rsync basics), but the token-security message is repeated four times: in the intro ("The agent never sees your authentication token..."), the Constants section (TOKEN_HANDLING), the setup section's 5-step list, and the Token Security defense-in-depth table plus behavioral rules — the setup steps and the guard table restate the same five protections nearly verbatim. This is more than anchor 4's 'minor instances that could be trimmed', but there is no concept padding or fluff, so it is above anchor 2. | 3 / 5 |
Actionability | Guidance is highly concrete — full bash blocks for pull/push/status, an exact rsync command with a complete exclude list, a per-hunk decision table with examples ("Lrage" for "Large"), and commit-message examples. However, the push and status blocks are not fully copy-paste ready: "cd paper-overleaf && git pull --ff-only" is followed by sibling-relative paths ("rsync ... paper/ paper-overleaf/" and "diff -rq --brief paper/ paper-overleaf/") that fail when run from inside paper-overleaf/, so anchor 5's 'fully executable' is not met. | 4 / 5 |
Workflow Clarity | Sequencing and validation are exemplary: a post-setup verification block with expected outcomes ("must show URL WITHOUT token", "must report Audit clean"), a pull-before-push rule, a per-hunk diff protocol that routes changes to re-run /paper-claim-audit or /citation-audit, an explicit confirmation gate before 'git push' to a shared resource, a three-way status decision table, and conflict-resolution rules that forbid destructive ops (no auto-merge, no reset --hard, no force push). Validation for the destructive/shared operations is present, so the workflow-clarity cap does not apply. | 5 / 5 |
Progressive Disclosure | The skill is a single ~210-line SKILL.md with no references/, scripts/, or assets/ directories, yet its two central technical guarantees (tools/overleaf_setup.sh and tools/overleaf_audit.sh) live in files not bundled with the skill, and detail-heavy content (the 5-layer defense-in-depth table, the per-hunk diff protocol) is inlined where a reference file would keep SKILL.md an overview. Structure within the file is good (clear headers, tables, See Also), but content that should be separate is inline, matching anchor 3 rather than anchor 4. | 3 / 5 |
Total | 15 / 20 Passed |