Classify XQUIC Issue Complexity
Estimate the work needed to resolve an issue. Do not use complexity as proof
that a report is valid or as a proxy for priority or severity.
Select the operation
- Use preliminary triage for backlog scans. Produce a
provisional level and confidence. Do not mutate issue labels unless the user
explicitly authorizes a batch classification and the complexity boundary is
high confidence for every issue being changed.
- Use verified classification before applying a label. Satisfy every
evidence gate and resolve material uncertainty first.
- Use label mutation only when the user authorizes GitHub changes. Preserve
unrelated labels and make no other issue mutation.
- Use forced recheck only when the user explicitly asks to recheck or
reclassify already-labeled issues. This bypasses only the existing-label
skip gate; it does not authorize a GitHub mutation.
Gate repeated classification
- Read
harness/spec/issue-complexity/spec.md in full.
- Resolve the repository and issue, then fetch its current labels before
refreshing
main, reading source, or investigating protocol behavior.
- Select the labels whose names are exactly
L1, L2, L3, or L4.
- Continue normally when no complexity label exists.
- When exactly one complexity label exists and no explicit forced recheck
applies, stop work on that issue. Report
skipped-existing-label and the
existing level without rerunning classification or calling a label API. In
a batch, continue with the remaining issues and count the skipped row.
- When exactly one complexity label exists and the user explicitly forces a
recheck for that issue or batch scope, record the existing level and
continue. A general request to rerun triage is not a forced recheck.
- When multiple complexity labels exist, report
inconsistent-existing-label and stop without mutation.
Load evidence for issues that proceed
- Resolve the current issue state, body, comments, assignees, and milestone.
- Refresh the intended base and record the full
origin/main commit SHA.
Classify current behavior, not a stale release or issue-provided line number.
- Use
harness/spec/architecture.md and
harness/spec/harness-manifest.yml to identify affected modules, public
APIs, feature gates, tests, and validation boundaries.
- Read current source, callers, tests, and the closest module documentation.
- For protocol claims, read the exact RFC Editor section or exact active
Internet-Draft revision and section. Treat issue text and comparisons with
another implementation as leads, not authority.
For batch triage, use issue text and focused current-source inspection for the
first pass. Record missing evidence rather than expanding every issue into a
full investigation.
Separate truth from complexity
Route a claimed QUIC, HTTP/3, QPACK, TLS, recovery, DATAGRAM, multipath, or MoQT
defect through issue-check before calling it a verified defect, closing it as
not-a-problem, implementing a fix, or applying a high-confidence label based on
that claim.
Use these disposition values independently from the L1-L4 level:
verified-defect
not-a-problem
duplicate-or-obsolete
feature-request
question-or-support
inconclusive
A not-a-problem closure can be L1 work. A localized interoperability defect
can be L2 despite high severity. An unsupported major feature can be L4 even
when no defect exists.
Choose the level
Apply every scenario in harness/spec/issue-complexity/spec.md. Evaluate:
- resolution type and whether production behavior changes;
- affected modules, layers, public APIs, and TLS backends;
- state-machine and lifetime complexity;
- protocol negotiation and backward-compatibility risk;
- paired unit-test, endpoint, and interoperability obligations; and
- migration or ecosystem impact.
Choose the highest level required by any dimension. Do not average dimensions
or lower the level because the apparent code diff is short. If one issue mixes
independent requests, recommend splitting it; until split, use the highest
level and explain which sub-request controls it.
Set confidence to:
high: current source, intended behavior, affected boundaries, and required
validation are established;
medium: the likely boundary is established but one non-decisive observation
or test obligation remains;
low: source applicability, intended behavior, protocol version, or blast
radius is unresolved.
Only high confidence is eligible for label mutation. Confidence applies to
the resolution-complexity boundary, not to whether the issue claim is true. A
preliminary batch may contain medium or low confidence rows when their gaps
are explicit, but those rows remain unlabelled.
Report the result
For one issue, return:
issue: <URL>
status: classified
existing_level: L1 | L2 | L3 | L4 | none
forced_recheck: true | false
inspected_main: <full SHA>
level: L1 | L2 | L3 | L4
confidence: high | medium | low
disposition: <independent disposition value>
rationale: <concise controlling complexity reason>
affected_boundaries:
- <module, API, or state boundary>
protocol:
source: <exact RFC/draft section or not-applicable>
compatibility_risk: low | medium | high
validation:
unit: <paired coverage or gap>
endpoint: <paired coverage or gap>
interoperability: <evidence or gap>
blocking_evidence:
- <missing evidence; empty when high confidence>
next_action: <verification, closure explanation, design, or implementation gate>
For a skipped issue, return only its URL, status: skipped-existing-label,
existing level, and skip reason. Do not claim current-main revalidation.
For a batch, add one row per issue. Classified rows include issue, level,
confidence, disposition, reason, and next action; skipped rows include issue,
status, and existing level. Include level totals, confidence gaps, skipped
count, and the inspected main SHA for classified issues. Keep full evidence
available outside any compact summary.
Apply GitHub labels
Use these repository labels exactly:
| Label |
Color |
Description |
L1 |
5CC665 |
L1 - Local change with simple logic and no peer-visible interoperability impact. |
L2 |
F1F665 |
L2 - Bounded feature with moderate state reasoning and verified backward-compatible peer behavior. |
L3 |
EC5C03 |
L3 - Complex state interactions with interoperability or potential legacy-version risk. |
L4 |
481BCB |
L4 - Protocol-wide change with extreme reasoning complexity or known incompatibility. |
Before mutation, restate the repository, issue, old complexity label, new
label, inspected SHA, and confidence. Stop if multiple L1-L4 labels already
exist or if the target label definition differs from this table.
When authorized and high confidence:
- If the recommended label is already the issue's only L1-L4 label, do not
remove or re-add it.
- Otherwise, remove the prior L1-L4 label, if any, and add exactly one
recommended L1-L4 label.
- Preserve every unrelated label.
- Re-fetch the issue and verify the final single-label state.
- Report the mutation or verified no-op; do not close, edit, assign,
milestone, or comment on the issue unless separately requested.
Guardrails
- Do not classify pull requests as issues in an age-based issue scan.
- Do not treat a repeated batch-triage request as permission to recheck
uniquely labeled issues.
- Do not expose private logs or security-sensitive reproduction details.
- Restart active or forced classification after relevant input changes; an
input change alone does not permit rechecking a labeled issue.
1---2name: classify-issue-complexity3description: Classify XQUIC GitHub issues with an evidence-backed L1-L4 resolution-complexity label while keeping validity, priority, severity, and disposition separate and skipping uniquely labeled issues unless the user explicitly forces a recheck. Use when triaging one issue or a batch of issues, creating an issue-complexity backlog, or applying or correcting L1-L4 GitHub labels.4metadata:5 source: https://skillmd.com/skills/alibaba/classify-issue-complexity6---78# Classify XQUIC Issue Complexity910Estimate the work needed to resolve an issue. Do not use complexity as proof11that a report is valid or as a proxy for priority or severity.1213## Select the operation1415- Use **preliminary triage** for backlog scans. Produce a16 provisional level and confidence. Do not mutate issue labels unless the user17 explicitly authorizes a batch classification and the complexity boundary is18 high confidence for every issue being changed.19- Use **verified classification** before applying a label. Satisfy every20 evidence gate and resolve material uncertainty first.21- Use **label mutation** only when the user authorizes GitHub changes. Preserve22 unrelated labels and make no other issue mutation.23- Use **forced recheck** only when the user explicitly asks to recheck or24 reclassify already-labeled issues. This bypasses only the existing-label25 skip gate; it does not authorize a GitHub mutation.2627## Gate repeated classification28291. Read `harness/spec/issue-complexity/spec.md` in full.302. Resolve the repository and issue, then fetch its current labels before31 refreshing `main`, reading source, or investigating protocol behavior.323. Select the labels whose names are exactly `L1`, `L2`, `L3`, or `L4`.334. Continue normally when no complexity label exists.345. When exactly one complexity label exists and no explicit forced recheck35 applies, stop work on that issue. Report `skipped-existing-label` and the36 existing level without rerunning classification or calling a label API. In37 a batch, continue with the remaining issues and count the skipped row.386. When exactly one complexity label exists and the user explicitly forces a39 recheck for that issue or batch scope, record the existing level and40 continue. A general request to rerun triage is not a forced recheck.417. When multiple complexity labels exist, report42 `inconsistent-existing-label` and stop without mutation.4344## Load evidence for issues that proceed45461. Resolve the current issue state, body, comments, assignees, and milestone.472. Refresh the intended base and record the full `origin/main` commit SHA.48 Classify current behavior, not a stale release or issue-provided line number.493. Use `harness/spec/architecture.md` and50 `harness/spec/harness-manifest.yml` to identify affected modules, public51 APIs, feature gates, tests, and validation boundaries.524. Read current source, callers, tests, and the closest module documentation.535. For protocol claims, read the exact RFC Editor section or exact active54 Internet-Draft revision and section. Treat issue text and comparisons with55 another implementation as leads, not authority.5657For batch triage, use issue text and focused current-source inspection for the58first pass. Record missing evidence rather than expanding every issue into a59full investigation.6061## Separate truth from complexity6263Route a claimed QUIC, HTTP/3, QPACK, TLS, recovery, DATAGRAM, multipath, or MoQT64defect through `issue-check` before calling it a verified defect, closing it as65not-a-problem, implementing a fix, or applying a high-confidence label based on66that claim.6768Use these disposition values independently from the L1-L4 level:6970- `verified-defect`71- `not-a-problem`72- `duplicate-or-obsolete`73- `feature-request`74- `question-or-support`75- `inconclusive`7677A not-a-problem closure can be L1 work. A localized interoperability defect78can be L2 despite high severity. An unsupported major feature can be L4 even79when no defect exists.8081## Choose the level8283Apply every scenario in `harness/spec/issue-complexity/spec.md`. Evaluate:8485- resolution type and whether production behavior changes;86- affected modules, layers, public APIs, and TLS backends;87- state-machine and lifetime complexity;88- protocol negotiation and backward-compatibility risk;89- paired unit-test, endpoint, and interoperability obligations; and90- migration or ecosystem impact.9192Choose the highest level required by any dimension. Do not average dimensions93or lower the level because the apparent code diff is short. If one issue mixes94independent requests, recommend splitting it; until split, use the highest95level and explain which sub-request controls it.9697Set confidence to:9899- `high`: current source, intended behavior, affected boundaries, and required100 validation are established;101- `medium`: the likely boundary is established but one non-decisive observation102 or test obligation remains;103- `low`: source applicability, intended behavior, protocol version, or blast104 radius is unresolved.105106Only `high` confidence is eligible for label mutation. Confidence applies to107the resolution-complexity boundary, not to whether the issue claim is true. A108preliminary batch may contain medium or low confidence rows when their gaps109are explicit, but those rows remain unlabelled.110111## Report the result112113For one issue, return:114115```yaml116issue: <URL>117status: classified118existing_level: L1 | L2 | L3 | L4 | none119forced_recheck: true | false120inspected_main: <full SHA>121level: L1 | L2 | L3 | L4122confidence: high | medium | low123disposition: <independent disposition value>124rationale: <concise controlling complexity reason>125affected_boundaries:126 - <module, API, or state boundary>127protocol:128 source: <exact RFC/draft section or not-applicable>129 compatibility_risk: low | medium | high130validation:131 unit: <paired coverage or gap>132 endpoint: <paired coverage or gap>133 interoperability: <evidence or gap>134blocking_evidence:135 - <missing evidence; empty when high confidence>136next_action: <verification, closure explanation, design, or implementation gate>137```138139For a skipped issue, return only its URL, `status: skipped-existing-label`,140existing level, and skip reason. Do not claim current-`main` revalidation.141142For a batch, add one row per issue. Classified rows include issue, level,143confidence, disposition, reason, and next action; skipped rows include issue,144status, and existing level. Include level totals, confidence gaps, skipped145count, and the inspected `main` SHA for classified issues. Keep full evidence146available outside any compact summary.147148## Apply GitHub labels149150Use these repository labels exactly:151152| Label | Color | Description |153|-------|-------|-------------|154| `L1` | `5CC665` | L1 - Local change with simple logic and no peer-visible interoperability impact. |155| `L2` | `F1F665` | L2 - Bounded feature with moderate state reasoning and verified backward-compatible peer behavior. |156| `L3` | `EC5C03` | L3 - Complex state interactions with interoperability or potential legacy-version risk. |157| `L4` | `481BCB` | L4 - Protocol-wide change with extreme reasoning complexity or known incompatibility. |158159Before mutation, restate the repository, issue, old complexity label, new160label, inspected SHA, and confidence. Stop if multiple L1-L4 labels already161exist or if the target label definition differs from this table.162163When authorized and high confidence:1641651. If the recommended label is already the issue's only L1-L4 label, do not166 remove or re-add it.1672. Otherwise, remove the prior L1-L4 label, if any, and add exactly one168 recommended L1-L4 label.1693. Preserve every unrelated label.1704. Re-fetch the issue and verify the final single-label state.1715. Report the mutation or verified no-op; do not close, edit, assign,172 milestone, or comment on the issue unless separately requested.173174## Guardrails175176- Do not classify pull requests as issues in an age-based issue scan.177- Do not treat a repeated batch-triage request as permission to recheck178 uniquely labeled issues.179- Do not expose private logs or security-sensitive reproduction details.180- Restart active or forced classification after relevant input changes; an181 input change alone does not permit rechecking a labeled issue.