catalan-adobe/memory-triage
Review Claude Code auto memory accumulated during a task and promote valuable findings to shared project config (CLAUDE.md, .claude/rules/, or global CLAUDE.md). Use this skill when finishing a branch, completing a task, wrapping up work, or anytime you want to review what Claude learned and decide what the team should know. Also use when the Stop hook reminds you about pending memory entries. Triggers on "triage memory", "review memory", "promote memory", "memory triage", "what did we learn", "clean up memory", "share learnings", "anything worth keeping", or when the user asks about knowledge accumulated during a session.
79
99%
Does it follow best practices?
Impact
—
No eval scenarios have been run
Passed
No known issues
Quality
Discovery
100%Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.
This is an excellent skill description that clearly defines a specific niche (memory triage and promotion), provides comprehensive trigger terms, and explicitly addresses both what the skill does and when to use it. The description is well-structured with concrete actions, specific file targets, and natural language triggers that users would actually say.
| Dimension | Reasoning | Score |
|---|---|---|
Specificity | Lists multiple specific concrete actions: review auto memory, promote findings to shared project config (CLAUDE.md, .claude/rules/, or global CLAUDE.md). Also specifies the context of when this happens (finishing a branch, completing a task, wrapping up work). | 3 / 3 |
Completeness | Clearly answers both 'what' (review auto memory and promote valuable findings to shared config files) and 'when' (explicit 'Use this skill when...' clause with multiple trigger scenarios, plus a 'Triggers on' list of specific phrases). | 3 / 3 |
Trigger Term Quality | Excellent coverage of natural trigger terms including 'triage memory', 'review memory', 'promote memory', 'what did we learn', 'clean up memory', 'share learnings', 'anything worth keeping', plus contextual triggers like finishing a branch or completing a task. These are terms users would naturally say. | 3 / 3 |
Distinctiveness Conflict Risk | Highly distinctive niche focused specifically on Claude Code's auto memory triage and promotion workflow. The specific mention of CLAUDE.md, .claude/rules/, memory entries, and the Stop hook make it very unlikely to conflict with other skills. | 3 / 3 |
Total | 12 / 12 Passed |
Implementation
92%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is a well-crafted skill with clear, sequential workflow steps, strong actionability through specific commands and paths, and excellent conciseness that respects Claude's intelligence. The main weakness is that it references two supporting files (CLASSIFICATION.md and OUTPUT-FORMATS.md) that are not provided in the bundle, meaning key details about classification criteria and output formatting are missing. The safety constraints (read-only memory, no auto-commits, secret redaction) are well-articulated.
Suggestions
Provide the referenced bundle files `references/CLASSIFICATION.md` and `references/OUTPUT-FORMATS.md`, or inline the essential classification criteria and output format examples directly in the SKILL.md if the bundle won't be available.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every section is lean and purposeful. No unnecessary explanations of what memory is or how Claude Code works. The encoding algorithm, classification heuristics, and promotion targets are all novel information Claude wouldn't inherently know. The important notes section is tight and each bullet earns its place. | 3 / 3 |
Actionability | Provides concrete, executable steps: specific shell commands (`git rev-parse --show-toplevel`), exact path construction algorithm, specific file paths (`~/.claude/projects/<encoded-path>/memory/`), exact secret patterns to match, and precise target file locations. The batch decision support and fallback behaviors are clearly specified. | 3 / 3 |
Workflow Clarity | Five clearly sequenced steps with explicit validation checkpoints: Step 1 has an early exit if no memory found, Step 3 includes secret redaction before presenting to user, Step 3 has a fallback for no candidates, Step 4 includes the 'show filtered' escape hatch, and Step 5 explicitly states read-only constraints and no auto-commits. The feedback loop of promote/discard decisions keeps the user in control. | 3 / 3 |
Progressive Disclosure | References to `references/CLASSIFICATION.md` and `references/OUTPUT-FORMATS.md` are well-signaled and one-level deep, which is good structure. However, no bundle files were provided, meaning these references are broken — the actual classification examples and output formats are missing, leaving gaps in the skill's completeness. The main body is well-organized but depends on content that doesn't exist. | 2 / 3 |
Total | 11 / 12 Passed |
Validation
100%Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.
Validation — 11 / 11 Passed
Validation for skill structure
No warnings or errors.
Reviewed
Table of Contents