Content
81%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 dense, well-engineered orchestration skill: the interpreter/gate/state contract is explicit and validated, routing is per-entry-point concrete, and detail is properly pushed into a verified one-level-deep bundle. The main costs are token weight and comprehension load in the Gate 2 and Contextual-offers sections, and a Files table that incompletely enumerates the actual bundle.
Suggestions
Trim or relocate the inline Contextual offers rules (host-filling table, cadence caps, session-state caveats) — they duplicate ../contextual-offers-for-startups/SKILL.md and consume ~15% of the body; keep only the inline minimum plus the pointer.
Split the 17-line Gate 2 paragraph into a short condition list or a small table (entry point × migration_plan state → Gate 2 offered?), moving edge cases like the build_deploy 'in_progress' fallback into migration-plan.md where they are acted on.
Complete the Files table: replace the 'references/decision-refs/*.md' wildcard with named rows and add the missing bundle files (the model-recommend phase directory and the *-assemble.md fragment/assembler files) so navigation does not depend on the frontmatter contract.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence — no AWS or concept tutorials, and nearly every line is skill-specific contract (e.g. the state-file JSON, the phase gate conditions) — so it is far above the verbose anchors. It stops short of 5 because the "Contextual offers" section inlines another skill's full rules (host tables, per-five-message caps, session-state caveats) while also pointing to "../contextual-offers-for-startups/SKILL.md for the full rules", and the Gate 2 paragraph packs deeply nested edge cases into a single block that could be trimmed or pushed to phase files. | 4 / 5 |
Actionability | Guidance is concrete and executable where it counts: "Check: `uv --version`", exact status conditions ("BOTH `phases.clarify == \"completed\"` AND `phases.confirm == \"completed\"`"), exact entry-point routing per flow, and a copy-paste state-file shape. Not 5: a few mechanisms stay prose-level rather than executable (the "${CLAUDE_PLUGIN_ROOT} does not resolve" fallback describes a condition but no concrete command), and most step-level execution is delegated to the reference files, leaving minor gaps in the body itself. | 4 / 5 |
Workflow Clarity | The sequence is explicit ("intake → discover → clarify → confirm → design → estimate → generate → migration-plan → poc → complete") with cold-start/warm-start entry rules, machine-checkable gates (Gate 1 / Gate 2 conditions on `recommendation_reviewed` and `phases.poc`), a read-merge-write state protocol, and error-recovery/resumability behavior ("if the session breaks between the \"yes\" and the load, the interpreter re-enters `poc` without re-asking"). Validation checkpoints are explicit throughout, matching the top anchor. | 5 / 5 |
Progressive Disclosure | Structure is genuinely strong: the Files table maps verified bundle files (phases, decision-refs, runtimes JSONs, scripts) to purposes, references are one level deep and well signaled ("Load it first"), and content is appropriately split (orchestration here, phase detail in phase files). Kept at 4 rather than 5 for minor organization gaps: the table uses a wildcard row ("references/decision-refs/*.md") and omits files that exist in the bundle (the `model-recommend` phase directory, `-assemble.md` fragment/assembler files), and the closing step points outside the skill root to unverifiable sibling paths ("../knowledge-base-for-startups/references/offers.md", "../architect-for-startups/references/observability.md"). | 4 / 5 |
Total | 17 / 20 Passed |