Content
85%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 process skill: ordered steps with genuine validation and feedback loops, appropriately split references that exist and stay one level deep, and concrete commands and file keys throughout. The only deductions are minor — a handful of motivational flourishes that could be cut, and the run/backtest mechanics being entirely reference-bound rather than partly executable inline.
Suggestions
Trim rhetorical flourish sentences (e.g. "whatever it looks like on the page", "and it is the one people skip", and the opening line of the Procedure section) to plain directives — they are the only padding in an otherwise lean body.
Inline one minimal executable selector example in step 5 (e.g. the single-lens `tessl code review` invocation from references/validation.md) so the core run step is actionable without opening the reference.
State inline what the two directed checks in step 6 compare (the backtest categories) in one line, so a reader can sanity-check backtest output before diving into references/validation.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is efficient and assumes competence — it explains only Tessl-specific mechanics ("Severity is decided after the lens runs", "--skill replaces the defaults rather than adding to them") and never re-explains known concepts. A few rhetorical flourishes could be trimmed, e.g. "whatever it looks like on the page", "and it is the one people skip", and "Going straight to the draft produces a lens that reads well and has never been shown to catch anything", which matches the score-4 anchor (minor instances that could be trimmed) rather than the every-token-earns-its-place anchor at 5. | 4 / 5 |
Actionability | Concrete guidance is present — copy-pasteable commands ("tessl code review --skill ./review-lenses/review-test-reliability", "tessl code review --skill your-workspace/your-plugin@1.0.0#review-test-reliability", "tessl whoami"), exact profile keys ("lenses", "globs", ".tessl-plugin/plugin.json"), and a concrete bar example ("whether a new test can fail for reasons unrelated to what it asserts"). The core run and backtest steps defer their executable mechanics to references/validation.md rather than including any inline selector example, leaving a minor gap consistent with the score-4 anchor. | 4 / 5 |
Workflow Clarity | The seven-step procedure is explicitly ordered ("Work through the steps in order") with validation checkpoints throughout: an agreement checkpoint in step 1, positive and negative test runs in step 5, a backtest comparison in step 6 with "Tune and re-run until the false positives and misses are ones you would ship. Report counts, not impressions", and a re-backtest requirement before adopting ("Re-backtest against the same set before adopting, so a gain in one category is not paid for out of another"). The tuning section maps each failure category to a specific edit, providing the feedback-loop structure of the top anchor. | 5 / 5 |
Progressive Disclosure | The bundle structure is clean: SKILL.md holds the procedure and platform facts, while the bulky details are split into two real one-level-deep references, each clearly signaled with its purpose and timing ("Read [references/lens-anatomy.md](references/lens-anatomy.md) first, for how far to direct a reviewer, a worked example, and what to keep out"). The references cross-link only to each other (siblings, not nesting), and the external ../code-review pointers are clearly signaled with section anchors — matching the top anchor for well-signaled one-level-deep references. | 5 / 5 |
Total | 18 / 20 Passed |