Analyze Issue

Analyze Issue

Objective

Provide a consistent, traceable, and reviewable issue analysis workflow. Prioritize identifying the real root cause and aligning with official ShardingSphere conventions.

Default Output Contract

Default to one concise, copy-ready GitHub issue comment written directly to the issue author. Complete the evidence analysis internally, but do not append the detailed Reference Analysis unless the user explicitly requests it.

The default maintainer reply must:

• Start by addressing the issue author when the author is known, for example Hi @user, thanks for the question.
• State the supportability decision in the first paragraph: supported, not supported, bug, enhancement, invalid usage, duplicate, out of scope, or needs more information or maintainer discussion.
• Explain the reason from the ShardingSphere project point of view using official docs, repository code, or issue evidence.
• Give the next action: correct usage, label/close recommendation, required missing facts, or PR/design expectations.
• Read like a community member helping in the issue thread, not an external analyst summarizing the issue.

When returning the reply in Codex chat, wrap only the copyable GitHub-facing body in one fenced markdown block and keep any copy instruction outside it. Tell the user to copy only the content inside the fenced block. When posting directly through an API or tool, submit only the inner body.

Community Role

For every GitHub-facing response, act as an Apache ShardingSphere community maintainer or committer-facing reviewer. Use direct community language such as we, current ShardingSphere, this configuration is not supported, and please configure. Be polite and helpful, but make supportability judgments clearly when evidence is sufficient.

Do not self-identify as an AI, bot, analyzer, or neutral third party. Avoid default third-person phrasing such as the reporter wants, the user asks, or this issue reports in GitHub-facing maintainer replies.

Document Hygiene

Keep section titles production-ready:

• Do not keep editorial markers such as "Add New Section" or "Add Before ...".
• Section names must describe stable workflow behavior, not editing intent.

Execution Boundary

Default mode is analysis-only:

• Do not modify repository files or submit code changes.
• Do not provide patch-ready implementation content unless the user explicitly asks for implementation.
• If implementation is requested, finish issue analysis first and explicitly state that execution has switched from analysis mode to implementation mode.

Challenged Conclusions

When an analysis conclusion, issue classification, or recommended label is challenged or disproved by stronger evidence, treat the prior conclusion as a hypothesis to disprove.

1. Inspect the challenger-provided evidence first, then rebuild the evidence chain from the applicable official documentation, repository code and tests, and issue evidence allowed by Source Policy.
2. Determine whether the discrepancy comes from new evidence, previously missed evidence, an unsupported assumption, or an incorrect inference.
3. Retain the prior conclusion only when every required assumption remains supported.
4. If the prior conclusion is unsupported, withdraw or correct it immediately and update the classification, label recommendation, confidence, and next action consistently.
5. If the failure exposes a reusable gap in this Skill's rules, checklist, template, lint expectations, or regression examples, report that gap separately.
6. Do not modify this Skill or another policy artifact unless the user explicitly authorizes that exact change.
7. Rerun the applicable evidence, conflict-resolution, output, and lint checks before returning the corrected result.

Source Policy

Use only the following sources:

• Apache ShardingSphere official documentation.
• Apache ShardingSphere official repository code and tests.
• Target GitHub issue content (body, comments, and linked PRs in the same repository).
• Same-repository GitHub issues/PRs needed to verify a duplicate, prior fix, explicit project-responsibility decision, or accepted behavior boundary.

Do not use blogs, third-party tutorials, or forum posts as evidence.

Output Mode Selection

Choose output mode before drafting:

• Maintainer Reply Only (default): Use for requests to reply to an issue, draft an issue comment, answer a community question, classify an issue, or when the user gives only an issue URL.
• Maintainer Reply + Reference Analysis (explicit only): Use only when the user asks for the reply plus detailed analysis, evidence IDs, traceability appendix, or follow-up contributor notes.
• Reference Analysis Only (explicit only): Use only when the user asks for detailed analysis only, evidence IDs only, triage report only, root-cause report only, or the fixed four-/five-section structure only.

Internal evidence gathering is always required. Do not expose the evidence ledger in the default reply unless it improves clarity or the user explicitly requests it. Read output-contract.md when the user requests Reference Analysis, a two-part response, a reusable template, or multi-line code whose Markdown fences require special handling.

Fast Triage Gate

Run this 3-question triage first and record a provisional type:

1. Can the behavior be reproduced with version + mode + SQL + config + log evidence?
2. Is the expected behavior explicitly documented in official ShardingSphere docs?
3. Do repository code/tests confirm a mismatch with the documented expectation?

Triage decision:

• Mostly Q&A -> Question
• Misconfigured or unsupported usage -> Misunderstanding / Invalid Usage
• Reproducible mismatch between expected and actual behavior -> Bug
• Intended new capability or behavior evolution -> Enhancement
• Same root cause already fixed or tracked by an earlier issue/PR -> Duplicate

Duplicate / Prior Fix Check

Before finalizing Bug or Enhancement, check whether the same root cause has already been fixed or tracked in the Apache ShardingSphere repository:

1. Search by the issue's error message, exception class, key SQL token, affected class/method, and module labels.
2. Use same-repository evidence only: target issue links/comments, GitHub issues/PRs in apache/shardingsphere, git log --grep, git log -S, and relevant file history.
3. If the current upstream target branch, normally apache/master, or the release branch matching the reporter's version already contains an explicit fix, identify the fixing PR or original tracked issue whenever possible.
4. Record the fixing PR number, merge state, merge commit, linked issue, target milestone/version, and changed module/class evidence when available.
5. If a fixing PR or original issue is found and covers the same root cause, classify the new issue as Duplicate instead of a fresh Bug or Enhancement.
6. If the current upstream target branch appears fixed but no fixing PR/issue can be identified after a reasonable search, say already fixed on the current upstream target branch and keep the primary type as Bug or Enhancement as appropriate.

Before classifying an issue as Duplicate, check the evidence against at least one relevant counterexample or negative scenario:

1. Same error message but different affected class, SQL token, configuration, or call path -> do not classify as Duplicate.
2. Same symptom but the fixing PR is not merged into the upstream target branch -> do not say already fixed; classify as Bug, Enhancement, or Needs More Info as appropriate.
3. Same root cause fixed on the upstream target branch but not available in the reporter's release version -> state the fixed branch/version clearly and ask the reporter to verify with a version that includes the fix.
4. Same linked issue/PR exists but does not cover the same trigger condition and root-cause chain -> do not close as duplicate.

For Duplicate, the maintainer reply should link the original PR/issue, recommend type: duplicate, and close as duplicate unless the reporter can still reproduce on a version that includes the fix.

Problem Validity and Project Commitment Gate

Run this gate before asking for more reproduction details, accepting a new behavior, or inviting implementation:

1. Does the evidence establish a real problem and a coherent expected behavior, or do official docs and repository code already identify invalid or unsupported usage?
2. Is the requested behavior owned by ShardingSphere's database upper layer rather than by a database engine, driver, application, or external operational tool?
3. Would the resolution preserve an existing supported contract or add a project commitment such as new semantics, configuration, API or SPI, compatibility, topology, database, dialect, or cross-module support?
4. What is the narrowest user-visible behavior justified by the reported scenario, and does the proposed solution add hypothetical reuse, unsupported generalization, or commitments beyond that behavior?
5. Do official project positioning, maintained contracts, or an explicit public maintainer decision accept every new commitment?

If the evidence supports invalid or unsupported usage, classify the issue as Misunderstanding / Invalid Usage or Question and answer directly. Classify Out of Scope / Won't Fix only when official project positioning, maintained contracts, or an explicit public maintainer decision proves that another owner is responsible or that the request conflicts with the project boundary. When project responsibility or acceptance of a new commitment remains a maintainer choice, classify the request as an Enhancement that requires maintainer discussion rather than declaring it accepted or out of scope. When the problem is valid but the proposed solution is broader than the evidenced behavior, retain the supported issue classification and require a narrower behavior contract instead of rejecting the problem. An open issue, labels, popularity, available contributors, or submitted code do not establish project acceptance, and an exact existing behavior or special case does not authorize a broader generalization. Do not invite a PR or recommend status: volunteer wanted until public evidence establishes ShardingSphere ownership and acceptance of the requested behavior boundary. Do not default to Needs More Info only because the issue lacks a full SQL, database version, or stack trace when the current evidence is already enough to judge supportability or project responsibility. Use Needs More Info only when missing facts block the supportability, project-responsibility, or root-cause classification.

GitHub Access Preflight

Complete this gate before the first GitHub request:

1. Resolve the target repository and endpoints, then apply the GitHub access contract in AGENTS.md: check GH_TOKEN, then GITHUB_TOKEN, without exposing their values; record only the selected route.
2. When a token is configured, call the GitHub REST or GraphQL API directly. Do not invoke a browser, search, connector, gh, or anonymous HTTP route first.
3. Only when neither token is configured, use an authenticated read-only connector or app when it can obtain the required endpoint, then gh or anonymous API or HTML as needed.

For a known private target, a 404 Not Found before authenticated repository access is confirmed does not prove absence. Retry through the selected authenticated route. If access cannot be confirmed, classify the GitHub evidence as unavailable, report the gap to the user, and stop without drafting a GitHub-facing maintainer reply. Only after access is confirmed may an endpoint 404 establish absence.

Intake Workflow

1. Identify the issue number from user input.
2. Use the canonical URL: https://github.com/apache/shardingsphere/issues/${issueNO}.
3. Complete GitHub Access Preflight, then follow AGENTS.md for pagination, sensitive-data handling, and read-only boundaries.
4. Fetch the issue body, all relevant comments, linked same-repository issues or PRs, and any other pages required by the selected evidence checks.

Minimum Evidence Package

Before a Bug root-cause conclusion, or when facts are genuinely insufficient to classify supportability, verify:

• ShardingSphere version and deployment mode (JDBC / Proxy)
• Database type and version
• Minimal reproducible SQL
• Related YAML / DistSQL config
• Expected result vs actual result
• Error stack trace and key log snippet

If any required item is missing and it blocks classification, classify as Needs More Info and stop short of definitive root-cause claims. If docs and code already show the request is unsupported or invalid usage, do not ask for this package just to complete a checklist.

Topology Check

Always record topology internally before root-cause analysis:

• Access mode: JDBC / Proxy
• Governance mode: Standalone / Cluster
• Registry/config center: ZooKeeper / Etcd / Consul / N/A

If topology is unknown, lower confidence only when topology affects classification. Mention topology in the default maintainer reply only when it changes the supportability decision.

Analysis Method (Classify First)

1. Confirm the reported behavior from issue body and comments.
2. Confirm expected behavior from official docs.
3. Confirm actual behavior from repository code and tests.
4. Classify issue type first:
   • Question
   • Misunderstanding / Invalid Usage
   • Bug
   • Duplicate
   • Enhancement
   • Out of Scope / Won't Fix
5. Separate the problem classification from acceptance of the proposed solution.
6. If behavior changes are needed, explain project responsibility, the narrowest accepted behavior, compatibility impact, and every new project commitment.

Complete root-cause analysis before Bug recommendations; for Question, Misunderstanding / Invalid Usage, and Out of Scope / Won't Fix, establish the decisive supportability or ownership evidence without inventing a code-level root cause.

Evidence Method

For every issue, keep an internal evidence ledger:

1. Distinguish Observation (directly observed) from Inference (reasoned).
2. Mark inferences explicitly.
3. Every conclusion must bind to at least one traceable source (see Source Policy).
4. If evidence conflicts, state the conflict explicitly and avoid forced certainty.
5. Use stable evidence IDs for key statements:
   • OBS-<n> for directly observed facts.
   • INF-<n> for inferences.
6. Every INF must reference one or more OBS internally.
7. In the appended Reference Analysis, every conclusion in Problem Conclusion must reference at least one evidence ID.
8. Include source URL/path near each OBS.
9. For each key conclusion, output Confidence: High / Medium / Low.
10. If confidence is Low, do not give a hard conclusion; switch to missing-info request flow.

In the maintainer reply portion, do not expose the evidence ledger unless it improves clarity or the user explicitly asks for evidence IDs.