Content
57%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 highly actionable, command-dense body whose escalation decision table and confirmation criteria are exemplary, but it is weakened by redundancy across four overlapping blind-SQLi sections, a batch extraction blueprint with no completion validation, and a total absence of progressive disclosure into reference files.
Suggestions
Consolidate the four blind-SQLi sections (Fast-Path, Manual-vs-Tool Decision Rule, Bulk Extraction Strategies, Parallel Blueprint) into a single escalation workflow — the same 'do not iterate manual probes after confirmation' rule is stated three separate times.
Add a completion-validation step to the parallel extraction blueprint: after reconstruction, assert that all STR_LEN position files exist and the result length equals STR_LEN before treating the secret as extracted.
Move the DBMS-specific manual payloads and the parallel extraction script into references/ files (e.g., references/payloads.md, references/blind-blueprint.sh), keeping SKILL.md as a lean overview with one-level-deep pointers.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is command-dense with no conceptual padding ('sqlmap -u ... --batch --dbs', payload blocks), but the blind-SQLi discipline is restated across four overlapping sections — 'Do NOT iterate manual character extraction', 'the next bash call MUST escalate to sqlmap', 'HARD RULE: ... not a single-character sequential probe' — that could be consolidated. Fits 'mostly efficient but could be tightened' (3) better than 'minor trims' (4). | 3 / 5 |
Actionability | Mostly executable guidance — the sqlmap invocations, hash-cracking loop, and the parallel extraction blueprint are near copy-paste ready with concrete parameters (--technique=BT --threads=5). Not 5: the '<KNOWN_TRUE_MARKER>' placeholder requires establishing a true-response baseline that is never demonstrated, and the blueprint's inner 'seq 32 126' sequential loop contradicts the binary-search strategy taught in the section above it. | 4 / 5 |
Workflow Clarity | Sequences are clear with explicit confirmation criteria ('time delay ≥5s OR boolean diff in response') and a state→action decision table including 'Submit/verify the partial result before launching a second harness'. However, the flagship batch workflow (20-concurrent-job parallel extraction) lacks completion validation — the reconstruct step globs pos_*.txt and silently skips missing positions, so a dropped request yields a short/garbled string undetected; per the batch-operations cap this holds at 3 rather than 4. | 3 / 5 |
Progressive Disclosure | Section headers are clear and well-ordered, but this is a ~260-line monolith with zero external files — the DBMS-specific manual payloads and the 45-line parallel extraction blueprint script are inline content that belongs in separate reference files. Fits 'some structure but content that should be separate is inline' (3); not 2 since organization within the file is good, and not 4 since there is no reference split at all for a file this long. | 3 / 5 |
Total | 13 / 20 Passed |