Conciseness

The body is mostly efficient — the 'How Hindsight Works' section teaches genuinely non-obvious server behavior (retain runs an extraction pipeline, so pass raw content), and commands come with tight, varied examples. But there is clear tightening headroom: the 'When to Store Memories' category list and the 'Best Practices' list repeat each other ('Attribute individual preferences', 'Include outcomes', rich-context advice appear twice), and 'Architecture decisions and their rationale' is literally duplicated under both 'Learnings from Tasks' and 'Team Knowledge'. Six retain examples plus a transcript example is more than needed. Matches anchor 3 (mostly efficient, some unnecessary explanation / could be tightened); not 2 because nothing explains concepts Claude already knows, and the padding is duplication rather than tutorial prose.

Actionability

Guidance is concrete and near copy-paste ready: exact bash commands for retain/recall/reflect with realistic fully-written payloads, the --context flag demonstrated with real values ('procedures', 'learnings', 'preferences'), and a complete setup recipe (curl install, heredoc config with chmod 600, bank-ID step). Matches anchor 4 (mostly executable, minor gaps): the <bank-id> placeholder must be substituted, and there is no indication of what recall/reflect output looks like or how to interpret it — small gaps that keep it from anchor 5.

Workflow Clarity

The core loop is clearly sequenced with an explicit validation checkpoint in setup ('cat ~/.hindsight/config' → 'If the file doesn't exist or is missing credentials, help the user set it up' → install → create config → get bank ID), and the operating rhythm is explicit: 'Always recall before' a enumerated list of situations, 'Store immediately' on discovery. Matches anchor 4 (clear sequence, most checkpoints present, minor validation gaps). Not 5: there is no feedback loop after retain (no check that the store succeeded or what the CLI prints on failure) and no recovery guidance if recall returns nothing; not 3 because validation is not absent — setup is explicitly verify-first.

Progressive Disclosure

No bundle files exist, so everything is inline in a single ~150-line SKILL.md with clear, well-ordered sections (Setup → How it works → Commands → When to store → When to recall → Best practices) and no nested or dead references. This is good structure for an instruction-only skill (per the rubric's code_vs_instruction note, absence of external files is not penalized per se), though the long taxonomy of memory categories, the six retain examples, and the transcript example are exactly the kind of content that could be offloaded to a reference file to slim the always-loaded body. Matches anchor 4 (good structure, most content appropriately placed, minor organization gaps); not 5 because the file carries reference-grade detail inline.

Specificity

The description names the domain (team knowledge/project conventions) with two concrete actions — 'Store team knowledge, project conventions, and learnings' and 'recall context before new tasks' — plus the deployment fact 'Connects to a self-hosted Hindsight server'. It sits between anchor 3 (1-2 concrete actions, not comprehensive) and anchor 4: the actions are concrete but few and generic (store/recall), with no finer-grained capability detail. Not level 2 because the actions are specific rather than minimal/generic; not clearly 5 because coverage is thin.

Completeness

Both parts are explicit: 'what' = 'Store team knowledge, project conventions, and learnings from tasks... recall context before new tasks', and 'when' = 'Use to remember what works and recall context before new tasks.' This satisfies the 'Use when...' requirement, so it clears the level-3 cap. It is not 5 because the 'when' clause is a single general condition ('before new tasks') without concrete trigger phrases like 'when the user mentions memory, asks to remember something, or references past decisions' — matching anchor 4 ('when' could be more explicit or specific).

Trigger Term Quality

Natural phrases present include 'team knowledge', 'project conventions', 'learnings', 'remember what works', and 'recall context' — genuinely likely user phrasing. However, common variations are missing: 'memory', 'save this', 'preferences', 'lessons learned', 'Hindsight'. Matches anchor 3 (relevant keywords but missing common variations/synonyms); not 4 because the synonym set is noticeably incomplete, not 2 because several natural (not jargon-only) terms appear.

Distinctiveness Conflict Risk

'Connects to a self-hosted Hindsight server' establishes a clear niche that separates it from generic memory or knowledge-base skills, and 'team knowledge'/'learnings' are reasonably distinctive triggers. Not 5: 'store knowledge' and 'recall context' still overlap with built-in memory or other knowledge-registry skills, so a minor overlap risk remains — matching anchor 4 (mostly distinct; minor overlap with closely related skills).