Content
92%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 excellent instruction-only skill body: highly actionable commands, workflows with real validation checkpoints, and clean progressive disclosure to on-disk and in-CLI references. The single notable weakness is redundancy — the runner-billing guidance is stated twice, and a few command-table entries are wordier than their value.
Suggestions
Merge the "Two runner-specific costs to keep in mind" paragraph in the Safety section into the "Interactive runners cost money" section (or reduce it to a one-line back-reference), since it repeats the same three points.
Tighten the longest command-table cells (e.g., `build delete`, `promotion resolve`, `run reattempt`) to their decision-relevant facts, leaving lifecycle detail to each command's `--help` per the skill's own overview-first policy.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious operational knowledge (billing, blind-retry hazards, environment resolution, the `--json` envelope exception) and never explains concepts Claude already knows, but the "Two runner-specific costs to keep in mind" paragraph in Safety restates the earlier "Interactive runners cost money" section (reuse one runner id, stop when done, run/act/exec unsafe to blind-retry), and several command-table cells run multi-sentence — minor trimming opportunities that keep it below "every token earns its place". | 4 / 5 |
Actionability | Concrete, copy-paste-ready commands cover the common cases (`qawolf agent send "$(cat prompt.md)"` with `--follow`, `qawolf agent send "<your answer>" --session <sessionId>`, `qawolf runner act --screenshot <path>`, `run get --flow-statuses failed`, `qawolf --json environment find`), and deferring per-command flag syntax to `qawolf <command> --help` is an explicitly stated, version-proof strategy rather than a gap in executable guidance. | 5 / 5 |
Workflow Clarity | Multi-step processes carry explicit validation checkpoints and feedback loops: git publishing (push, poll `environment get` until `lastSyncedCommitHash` includes the commit, then read `flowId`/`url`), agent sessions (a question prints and exits 0 — answer via `--session`, then re-follow, and read the final `status` from the JSON line rather than the exit code), and write safety (a successful write is confirmation, never blind-retry on timeout, read a trigger before `trigger update`). No validation gaps were found on destructive or batch paths. | 5 / 5 |
Progressive Disclosure | The body is explicitly an overview ("This file is an overview, not a reference") that points to two well-signaled, one-level-deep references — references/run-results.md and references/runner.md, both present in the bundle — with bolded "Read ... before ..." cues, fallback fetch URLs, and the in-CLI `qawolf help ref runner` guide; the large commands table functions as a navigational index rather than inlined reference detail. | 5 / 5 |
Total | 19 / 20 Passed |