Content
73%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, largely executable migration runbook whose strongest assets are explicit sequencing, operator-approval gates, thorough post-migration validation, and concrete error-recovery guidance. Its weaknesses are duplicated requirements across sections and per-source sections (Notion, CSV, JSON, Roam) that lack the concrete commands the company and Obsidian paths enjoy.
Suggestions
Deduplicate the safety rules: state the additive-only and no-bypass rules once (in the Contract or "When it fails" section) instead of repeating them across the company-repositories section, and merge the overlapping Phases and Verification steps.
Add a concrete put_page example for the generic path (e.g., one CSV row mapped to slug/frontmatter/compiled_truth) so Notion, CSV, and JSON migrations are as executable as the company-repository and Obsidian paths.
Move the "(v0.12.1+)" version pin out of the workflow prose into a version-compatibility note so time-sensitive details don't burden the main instructions.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is almost purely instructional with no conceptual padding, but requirements are repeated: the additive-only rule appears as both "Never fix the source files or commit dirty changes automatically" and the Anti-Pattern "Never modify, move, or delete the source files"; the global-schema bypass warning and "missing embeddings are expected" each appear twice; the Phases and Verification sections restate the sample-test and health-check steps; and a version pin "(v0.12.1+)" is embedded inline. This fits "mostly efficient but could be tightened" rather than anchor 4's minor-trimming level. | 3 / 5 |
Actionability | The company-repository and Obsidian paths give exact executable commands with flags ("gbrain sources connect <path> --brain <id> --source <id> --profile company-brain --json", "gbrain sync --brain <id> --source <id> --no-embed --no-pull", "gbrain extract links --source db --dry-run"), but the Notion/CSV/JSON/Roam sections offer only procedural steps like "Map columns to frontmatter fields" with no concrete command or put_page example. Mostly executable guidance with moderate gaps — above anchor 3, below anchor 5's copy-paste coverage of common cases. | 4 / 5 |
Workflow Clarity | Sequencing is explicit with validation gates and feedback loops throughout: "Test with a sample. Import 5-10 files, verify by reading them back", "Only after the operator approves that preview, rerun with --yes", a five-item verification checklist (page count, health, round-trip export, spot-check, search test), and a "When it fails" section with specific failure modes and a resume command. This is a batch operation whose validation is thorough, so no cap applies, and it matches anchor 5's explicit-checkpoint, error-recovery profile. | 5 / 5 |
Progressive Disclosure | Headers, a supported-sources table, and clearly signaled one-level references ("Read skills/conventions/brain-routing.md and skills/conventions/untrusted-content.md first", "Full behavior and errors: docs/guides/company-brain-ingestion.md") give good structure, but no bundle files exist to back the referenced paths, and the ~30-line company-repository procedure plus per-source guides are inlined rather than split into reference files. Good structure with minor organization gaps — below anchor 5's clean overview-plus-references split. | 4 / 5 |
Total | 16 / 20 Passed |