forked from Gitlink/gitlink-cli
6.3 KiB
6.3 KiB
GitLink CLI Agent Workflow Enhancement Suite
1. Background
GitLink CLI serves both human maintainers and AI Agents. The competition focuses on intelligent open-source contribution workflows, where structured analysis, stable output, and safe automation matter more than raw command count.
2. Problem
Open-source maintenance often suffers from:
- Issue backlog and delayed triage
- High PR review cost
- Repetitive release note preparation
- Lack of structured repository health evaluation
- AI Agents needing stable, machine-readable output
3. Solution
This project extends GitLink CLI with the GitLink CLI Agent Workflow Enhancement Suite.
Implemented now:
workflow +triageworkflow +healthworkflow +pr-summaryworkflow +repo-report- read-only GitLink fetch layer for workflow triage and health
- read-only PR metadata, changed files, and commits fetch layer for PR summary
- partial read-only repository report aggregation for health, issues, and PR list metadata
- expanded fetch boundary tests for empty responses, label and author normalization, error-in-body handling, alternative activity timestamps, release shapes, and CI unavailability
- local-first analysis with no LLM dependency
- stable Agent-facing JSON / table / markdown output
Planned next:
workflow +release-notesworkflow +stale
4. Technical Route
- Go + Cobra + existing shortcut architecture
- rule-based analysis
- stable DTOs
json/table/markdownrenderersen/zh-CNmessage mapping- no LLM dependency
- local-first, dry-run-safe workflow design
5. Implemented Features
| Command | Status | Main Value |
|---|---|---|
workflow +triage |
Done | Issue classification, priority, missing information, and actions |
workflow +health |
Done | Repository health score, risk level, and recommendations |
workflow +pr-summary |
Done | PR risk, review focus, test suggestions, and merge checklist |
workflow +repo-report |
Done | Aggregated repository workflow report for maintainers and Agents |
workflow +release-notes |
Planned | Release note generation from PR titles and commits |
workflow +stale |
Planned | Stale issue and PR analysis |
workflow +triage
- issue type detection
- priority scoring
- confidence scoring
- missing information detection
- risk flags
- recommended action
- suggested comment
- reasoning and matched rules
workflow +health
- health score
- risk level
- metrics
- scoring notes
- recommendations
- unknown metric tolerance
workflow +pr-summary
- change type detection
- risk level analysis
- review focus generation
- test suggestion generation
- merge checklist generation
- read-only fetch of PR metadata, changed files, and commits
workflow +repo-report
- one-command repository workflow report
- health, issue triage, and PR summary aggregation
- report score and overall risk level
- partial report behavior when optional remote sections fail
- markdown output for competition and maintainer reports
- JSON output for Agent consumption
6. Innovation Points
- Agent-native structured output
- rule-based intelligence without external LLM dependency
- explainable workflow decisions
- safety-first local analysis
- bilingual command output
- extensible workflow command design
- competition-friendly incremental PR path
7. Testing and Verification
- Unit tests cover triage, health scoring, messages, rendering, and command helpers.
- Fetch-layer tests cover issue normalization, repository health probing,
and PR metadata/file/commit normalization with
httptest. - Boundary tests cover empty responses, label and author normalization, error-in-body handling, alternative activity timestamps, release response shapes, and CI unavailability.
- PR summary tests cover docs-only, workflow code, internal client, security-sensitive, mixed-file, zh-CN, render, command, and fetch-failure cases.
- Repo report tests cover aggregation, scoring, JSON/table/markdown rendering, command wiring, local JSON input, partial fetch behavior, and include flags.
- Local command examples were executed successfully.
- Full repository testing passed in the current environment.
- Automated tests use
httptestand do not depend on real remote API availability.
8. Demonstration Plan
Official repository
Use Gitlink/gitlink-cli as the reference repository:
workflow +triagewith English table outputworkflow +triagewith security JSON outputworkflow +triagewith Chinese markdown outputworkflow +healthwith table outputworkflow +healthwith risky JSON outputworkflow +pr-summarywith markdown outputworkflow +repo-reportwith markdown output for the full competition story- Explain how agents consume stable JSON
Self-built test repository
Use a small demo repository to show:
- bug triage
- security triage
- docs triage
- healthy repo score
- risky repo score
- full repo report from
shortcuts/workflow/testdata/repo_report.json
9. Roadmap
- Phase 1: local workflow prototype, completed
- Phase 2: API fetch and normalization, completed
- Phase 3:
pr-summary, completed - Phase 4:
repo-report, completed - Phase 5:
release-notes,stale
10. PR Plan
- PR 1: workflow rule engine and local commands
- PR 2: documentation and tests
- PR 3: API fetch layer
- PR 4:
pr-summary - PR 5:
repo-report - PR 6:
release-notes/stale
11. Evaluation Mapping
| Criterion | Evidence |
|---|---|
| 功能完整性 20% | Four implemented commands cover Issue triage, health scoring, PR summary, and repo report |
| 创新性 20% | Agent-native JSON, explainable rules, local-first safety model, repository workflow report |
| 实用价值 20% | Reduces maintainer triage/review overhead and creates copy-ready markdown reports |
| 文档与演示 20% | README, design doc, test report, competition write-up, demo script, defense Q&A |
| 成果落地 20% | Prepared for GitLink official PR, CI verification, and maintainer review iteration |
12. Landing Plan
- Push the implementation branch to the public repository.
- Create a GitLink official PR against
Gitlink/gitlink-cli. - Record CI result and PR URL in
docs/final-submission-checklist.md. - Respond to maintainer review over the expected 1-2 week review cycle.
- Keep
release-notesandstaleas follow-up work instead of expanding this PR further.