java-heap-dump-triage

分析 Java heap dump、.hprof 文件、Eclipse MAT HTML 报告、Leak Suspects zip、Class Histogram、Dominator Tree、GC 异常、JVM OOM 或内存泄漏问题，并把 retained heap 证据关联回 Java/Kotlin/Spring/Quartz 项目代码，产出中文的根因判断、修复方案、验证计划和故障复盘文档。适用于用户要求分析 dump/hprof/heap dump/MAT 报告、内存泄漏、OOM、retained heap、dominator tree、RAMJobStore、ThreadLocal 泄漏、缓存泄漏、classloader 泄漏，或要求结合仓库代码优化内存问题的场景。

Quality

90%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

75%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is a strong, highly actionable operational guide: concrete commands with expected artifacts, a well-sequenced 9-step workflow with validation checkpoints, and a properly signaled one-level reference. Its main weaknesses are token efficiency — a run-log section duplicates an existing command and carries machine/user-specific details — and the absence of explicit failure-recovery branches in the workflow.

Suggestions

Trim or generalize the 实测生成流程记录 section: drop the repeated ParseHeapDump.sh command (already in 常用命令) and replace personal paths and byte sizes with the generalized expectations (exit code 0 + the three report zips present).

Move the run-log narrative and the more case-specific 诊断经验 items into a reference file (e.g., references/mat-report-generation.md), keeping SKILL.md as a lean overview that points to it.

Add explicit failure-recovery branches to the workflow: what to do if no MAT installation is found (install/download hint), if ParseHeapDump.sh exits non-zero, or if expected report zips are missing after generation.

Dimension	Reasoning	Score
Conciseness	Mostly efficient operational guidance, but the '实测生成流程记录' section repeats the ParseHeapDump.sh invocation already given in 常用命令 and includes instance-specific details ('/Users/greysonfang/Desktop/heap_dump.hprof', '2.6G', '168K/140K/772K') that add tokens without generalizable value. Not score 4 because the duplicated command and personal-path log are noticeable tightening opportunities; not score 2 because the rest is dense and free of concept explanations Claude already knows.	3 / 5
Actionability	Fully executable, copy-paste-ready commands cover the common cases: find patterns for dumps, 'command -v ParseHeapDump.sh', the full 'ParseHeapDump.sh <dump.hprof> org.eclipse.mat.api:suspects ...' invocation, a complete python3 heredoc for HTML text extraction, and a targeted rg query — plus expected artifact names and the Quartz class-name triggers. Not score 4 because no gaps remain: inputs, expected outputs, and verification criteria are all specified.	5 / 5
Workflow Clarity	The 9-step sequence is clear with real checkpoints: '以最终退出码和生成物为准' for report generation, a dedicated 验证 step (narrowest-scope compile/tests, re-dump and metrics online), and the rule to label insufficient evidence as '假设'. Not score 5 because explicit failure-recovery branches are thin — beyond the log-truncation note there is no 'if X fails, do Y' loop (e.g., MAT not found, report generation failing), so checkpoints are present but not full feedback loops.	4 / 5
Progressive Disclosure	The single bundle reference is real (references/quartz-ramjobstore.md exists), one level deep, and clearly signaled by matching class-name triggers ('RAMJobStore、CronTriggerImpl、TriggerWrapper... 读取 references/quartz-ramjobstore.md'). Not score 5 because material that plausibly belongs in references — the 实测生成流程记录 run log and part of 诊断经验 — is inlined in SKILL.md rather than split out, leaving the overview heavier than the anchor's cleanly split structure.	4 / 5
	Total	16 / 20 Passed

Description

100%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is exemplary: it names concrete capabilities (MAT report analysis, retained-heap-to-code correlation, deliverables), then gives an explicit use-when clause packed with natural trigger terms, synonyms, and file extensions, all in third person with no filler. It directly matches the strongest anchors on every dimension.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions with comprehensive coverage: '分析 Java heap dump、.hprof 文件、Eclipse MAT HTML 报告、Leak Suspects zip、Class Histogram、Dominator Tree' plus '把 retained heap 证据关联回 Java/Kotlin/Spring/Quartz 项目代码' and '产出中文的根因判断、修复方案、验证计划和故障复盘文档'. Not score 4 because coverage of the domain's actions (parse artifacts, correlate to code, produce deliverables) is comprehensive rather than having minor gaps.	5 / 5
Completeness	Explicitly answers both: 'what' — analyze dump/MAT artifacts, correlate retained heap evidence to code, and produce root-cause/fix/verification/postmortem outputs; 'when' — '适用于用户要求分析 dump/hprof/heap dump/MAT 报告…的场景', an explicit Use-when clause with concrete trigger phrases. Not score 4 because the 'when' is fully explicit with specific triggers, not merely adequate.	5 / 5
Trigger Term Quality	The 'when' clause enumerates natural user phrasings with synonyms and file extensions: 'dump/hprof/heap dump/MAT 报告、内存泄漏、OOM、retained heap、dominator tree、RAMJobStore、ThreadLocal 泄漏、缓存泄漏、classloader 泄漏'. Not score 4 because synonyms (dump/hprof/heap dump) and the concrete file extension (.hprof) are all present, matching the comprehensive anchor.	5 / 5
Distinctiveness Conflict Risk	Clear niche (Java heap dump triage via MAT) with distinct triggers like 'RAMJobStore、ThreadLocal 泄漏、classloader 泄漏、.hprof、Leak Suspects zip' that no adjacent skill (general JVM tuning, code review) would claim. Not score 4 because overlap risk is minimal, not merely minor.	5 / 5
	Total	20 / 20 Passed

Validation

100%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository: TencentBlueKing/bk-ci
Path: ai/skills/java-heap-dump-triage/SKILL.md
Commit: d5420f2

Reviewed: 9 days ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.