Content
27%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill demonstrates deep domain thinking about planning under uncertainty and has a well-defined vocabulary and output contract, but it is severely over-engineered for a context-window-resident instruction file. The exhaustive enumeration lists (Safe Default Gate, Ambiguity Escalation, Report Back categories) consume enormous token budget without proportional value — most items are judgment calls Claude can make from shorter guidance. The lack of a concrete worked example and the absence of progressive disclosure into bundle files are the most impactful weaknesses.
Suggestions
Reduce the Safe Default Gate from 16 criteria to 3-5 key principles with a summary heuristic (e.g., 'bounded, reversible, convention-following, non-user-visible, verifiable'), and move the full checklist to a reference file if needed.
Condense the 12 Ambiguity Escalation Rules and 15 Report Back categories into grouped summaries (e.g., 'security/compliance → HARD-BLOCKED', 'external side effects → ASK-USER') and move detailed enumerations to a bundle file like ESCALATION_RULES.md.
Add one concrete worked example showing a sample backlog item analyzed through the full workflow, producing a filled-in output contract with completeness summary, missing inputs table, and planning annotations.
Extract the workflow into an explicit numbered sequence (e.g., '1. Gather inputs → 2. Classify each → 3. Run Safe Default Gate where applicable → 4. Produce report → 5. Validate no invented facts') with clear checkpoints between phases.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is extremely verbose at ~400+ lines. It over-explains concepts Claude already understands (e.g., what context windows are, what ambiguity means), includes exhaustive enumeration lists (15 Safe Default Gate criteria, 12 ambiguity escalation rules, 15 report-back categories) that could be condensed, and repeats the same principles multiple times across sections. The complexity model explanation and mermaid diagram add tokens for a concept (knowledge-bounded task sizing) that could be stated in 2-3 sentences. | 1 / 3 |
Actionability | The skill provides concrete output structures (completeness summary statuses, missing input report fields, XML concerns block format) and specific classification vocabularies, which is good. However, it lacks executable examples — no sample input scenario with a worked-through analysis, no concrete example of a completed report, and the output contract sections describe fields without showing a filled-in example. The guidance is detailed but largely declarative rather than demonstrative. | 2 / 3 |
Workflow Clarity | The workflow is implicitly sequenced (analyze inputs → classify → produce report → annotate tasks) but never presented as an explicit ordered sequence with validation checkpoints. The behavioral rules section lists what the planner MUST do but doesn't provide a clear step-by-step execution flow. The recursive question discovery process is described but lacks a concrete stopping condition or validation gate. The reflection checkpoint is a good validation step but is buried in a bullet list rather than integrated into a clear workflow. | 2 / 3 |
Progressive Disclosure | The entire skill is a monolithic wall of text with no references to supporting files despite the content being long enough to warrant splitting. The Safe Default Gate (16 criteria), Ambiguity Escalation Rules (12 rules with routing), and Report Back categories (15 categories) could each be separate reference files. The sister skill reference to dh:rt-ica is mentioned but not linked. No bundle files exist to offload the extensive enumeration lists. | 1 / 3 |
Total | 6 / 12 Passed |