Content
52%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.
The operational core is excellent β concrete tools, commands, config examples, and clear procedures for storing, retrieving, correcting, and debugging memory β but it is buried in an ~83KB monolith. The entire configuration reference and several policy deep-dives are inlined with no bundle files at all, and several topics are explained two or three times, making token cost the skill's dominant defect.
Suggestions
Move the full configuration reference table and the 'Spreading refreshes with H', observation-scopes, and manageBankConfig/defaultBankConfig deep-dives into a references/ file (e.g. references/config.md), keeping SKILL.md to the operational core: automatic behavior, storing, retrieving, correcting, install, and quick diagnostics.
State each fact once: install/autoUpdate behavior is covered in both the 'Install / update' section and the reference table row, and page-refresh policy appears in the table, the prose after it, and the 'Spreading refreshes' section β pick one location per topic and cross-link.
Replace multi-hundred-word table cells with one-line summaries that link to the detailed section or reference file, so the reference table serves as an index rather than an essay.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is ~742 lines / ~83KB with config-table cells running to several hundred words each (e.g. the autoInject, recallOptions, and pages rows), and key topics are explained two or three times over (install/autoUpdate in both the 'Install / update' section and the reference table; page-refresh policy in the table, the prose after it, and the 'Spreading refreshes with H' section). It is noticeably verbose with padded sections, though it does not explain concepts Claude already knows, so it does not fall to anchor 1. | 2 / 5 |
Actionability | Concrete, executable guidance throughout: tool signatures like hindsight_ingest_document(title, content) and hindsight_capture_initiative(title, summary), copy-paste npx commands for install/update/stats, copy-paste JSON config examples, a verbatim crediting format ('π§ From Hindsight memory (<page>): β¦'), and a structured correction procedure specifying title and content layout. Minor gaps β tool usage is shown signature-style rather than as a worked invocation, and some tools (hindsight_diagnose, hindsight_sync_status) are named without usage detail β keep it below fully copy-paste-ready anchor 5. | 4 / 5 |
Workflow Clarity | The agent-facing workflows are clearly sequenced: 'Storing things deliberately' is a case-by-case decision list, 'Correcting wrong or stale memory' gives a fix-the-record procedure with an explicit verify-first precondition and required content structure, and 'When memory seems to be missing' points to specific diag events to check. Checkpoints are present for the risky operation (correcting memory only after verification; bank deletion consequences spelled out), with only minor gaps such as the config-edit application table being reference material rather than a workflow. | 4 / 5 |
Progressive Disclosure | No bundle files exist at all (no references/, scripts/, or assets/), and roughly half the body β the full configuration reference table plus the deep dives on page-refresh cron hashing, observation scopes, and manageBankConfig/defaultBankConfig β is inlined in SKILL.md where it clearly belongs in one or more separate reference files. Section headers are good, which lifts it above unstructured anchor 2's example, but the total absence of reference files and the scale of inlining (~400+ lines of config reference) place it at anchor 2 rather than 3, whose example at least signals an external ADVANCED.md. | 2 / 5 |
Total | 12 / 20 Passed |