SuperAPI AI Directory

git-precommit

Writing & research

View repository

Updated 10-16 07:58

About this skill

Pre-Git-Commit Workflow (Orchestrator + Pre-Commit Hooks)

Copilot Hooks (real-time corrections while writing) are defined in auto-paper/SKILL.md. Pre-Commit Hooks (this file, final gatekeeping before git commit) only report and do not auto-modify.

Triggers: "commit", "push", "done", "wrapping up", "quick commit"


Step 0: Detect change scope

get_changed_files() or git diff --cached --name-only. Paper file pattern:projects/*/drafts/*, projects/*/concept.md, projects/*/references/*, projects/*/.memory/* → Triggers Paper Hooks when matched.


General Hooks (triggered every time)

Hook Condition Action MCP Tools
G1 memory-sync Required Update Memory Bank + workspace state memory_bank_update_progress, sync_workspace_state
G2 readme-update New feature/API change Update README read_file, replace_string_in_file
G3 changelog-update Version/important change Update CHANGELOG Same as above
G4 roadmap-update Milestone completed Update ROADMAP Same as above
G5 arch-check Structural change Check architecture docs grep_search, list_dir
G6 project-integrity SKILL/AGENTS/src change Project consistency audit See below
G7 vsx-integrity skills/prompts/vsx change VSX Extension sync See below
G8 doc-update-reminder Any code/config change Remind about docs that need updating scripts/check-doc-updates.py (pre-commit hook)
G9 git-status-check Required Detect uncommitted/unpushed (Code-Enforced) run_writing_hooks G9

G6 Project Consistency Audit

# Check Method Failure
G6.1 Tool count grep -c "mcp.tool" vs README/ARCHITECTURE ⚠️ WARN
G6.2 Skill count ls -d .claude/skills/*/ vs AGENTS.md ⚠️ WARN
G6.3 Prompt count ls .github/prompts/*.prompt.md vs docs ⚠️ WARN
G6.4 Hook tool exists In SKILL.md mcp_mdpaper_* → registered ❌ FAIL
G6.5 Cross-file number consistency README vs ARCHITECTURE vs AGENTS ⚠️ WARN

G7 VSX Extension Sync

# Check Method Failure
G7.1 Skills sync diff source vs bundled ⚠️ WARN
G7.2 Prompts sync diff source vs bundled ⚠️ WARN
G7.3 Chat commands package.json chatParticipants ❌ FAIL
G7.4 Version semver Format validation ❌ FAIL
G7.5 TypeScript compilation tsc --noEmit ⚠️ WARN

Quick fix:cd vscode-extension && ./scripts/build.sh


Paper Hooks (when paper changes are detected)

Hook Check MCP Tools Verdict
P1 citation-integrity [[wikilinks]] parseable scan_draft_citations, validate_wikilinks 0 unresolved = ✅, unknown = ❌
P2 anti-ai-scan AI-trace wording read_draft + Agent scan 0 = ✅, 1-2 = ⚠️, ≥3 = ❌
P3 concept-alignment NOVELTY/SELLING POINTS reflected read_draft("concept.md") + drafts Complete = ✅, Partial = ⚠️, NOVELTY missing = ❌
P4 word-count Each section ±20% count_words Over 50% = ❌
P5 protected-content 🔒 Block exists and is non-empty read_draft("concept.md") Missing = ❌
P6 memory-sync .memory/ updated sync_workspace_state Not updated = AUTO-FIX
P7 reference-integrity Reference metadata complete list_saved_references, get_reference_details Not VERIFIED = ⚠️
P8 methodology Methods reproducible read_draft + Agent checklist Items <3 = ❌

P2 Anti-AI prohibited words:In recent years, It is worth noting, plays a crucial role, has garnered significant attention, a comprehensive understanding, This groundbreaking, delve into, shed light on, pave the way, a myriad of

P8 complements Copilot Hook B5: B5 corrects in real time while writing, P8 gives final confirmation at commit time (safety net).


Hook performance tracking (§23 Self-Improving)

Record after each execution: projects/{slug}/.audit/precommit-stats.md: trigger rate/pass rate/warning rate/block rate.

  • Pass rate >95% (5+ times) → possibly too loose
  • Block rate >50% (5+ times) → possibly too strict
  • Records for auto-paper Hook D analysis

Execution mode

Mode Trigger Execution
Standard "Prepare commit" Step 0 → G1-G7 → P1-P8 (if applicable) → Final
Quick "Quick commit" G1 + P1 + P5 (if applicable) → Final
Development 「commit code」 G1-G7 (skip Paper Hooks) → Final

Git operations

Commit Message format:type(scope): description

  • Types: feat, fix, docs, refactor, style, test, chore
  • Scope: paper, concept, refs, export, core

Skill dependencies

Hook Orchestration Skill Tools
G1 memory-updater memory_bank_update_progress
G2-G4 readme/changelog/roadmap-updater read_file, replace_string_in_file
G5 ddd-architect grep_search, list_dir
P1-P4 draft-writing read_draft, count_words, validate_wikilinks
P3,P5,P8 concept-development read_draft("concept.md")
P7 reference-management list_saved_references, get_reference_details