gitnexus-plan

Generates implementation-ready engineering plans from GitNexus knowledge graphs, PDG analysis, and source verification.

46.8k|5.1k|Updated Aug 2, 2025
One-click install
npx skills add https://github.com/abhigyanpatwari/GitNexus --skill gitnexus-plan
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: gitnexus-plan
Source: https://github.com/abhigyanpatwari/GitNexus/tree/main/gitnexus-claude-plugin/skills/gitnexus-plan
Command: npx skills add https://github.com/abhigyanpatwari/GitNexus --skill gitnexus-plan

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

AI coding agents often make blind edits because they lack architectural context about a codebase. This Skill produces deep, implementation-ready engineering plans by combining GitNexus repository graph intelligence, statement-level Program Dependence Graph analysis, and targeted source verification, so a follow-up implementation agent can start work without re-investigating the repository.

Core Features & Use Cases

  • Graph-guided planning: Uses GitNexus query, context, impact, and trace tools to map callers, callees, blast radius, and execution flows before proposing changes.
  • Statement-level PDG slices: Builds bounded control- and data-dependence slices for the functions central to a change, with dedicated security and performance analysis modes.
  • Machine-readable context pack: Emits a 13-section plan with a versioned evidence-provenance snapshot (dirty-tree digest and cited-path manifest) that an executor agent consumes directly.
  • Deepen mode: Strengthens an existing plan in place, re-verifying claims against a refreshed index and reconciling already-landed execution steps.
  • Use Case: Run "/gitnexus-plan Add retry support to the ingestion pipeline" to receive a verified plan document at docs/plans/ with direct dependents, test scenarios, risks, and verification commands.

Quick Start

Ask your agent to run /gitnexus-plan followed by a task description, such as planning a bug fix or feature, after indexing the repository with npx gitnexus analyze.

Frequently Asked Questions about gitnexus-plan

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate an implementation plan for a code change with GitNexus?

Run /gitnexus-plan followed by your task description in Claude Code, or ask any AGENTS.md-aware agent to follow the skill's SKILL.md. The skill classifies the task, navigates the GitNexus graph, verifies source, and writes a plan to docs/plans/.

What is the difference between gitnexus-plan compact and full plan forms?

Compact plans cover core sections in under 80 lines with a mini context pack, suited to bug fixes and small features. Full plans emit all 13 sections with impact depth 3 and PDG slices, used for refactors, security, performance, and architecture work.

Does gitnexus-plan work without a GitNexus index?

Yes, it falls back to targeted grep and read exploration, labeling all findings as source-derived and recommending an index. Statement-level PDG claims are skipped entirely when the PDG layer is unavailable rather than fabricated.

Can gitnexus-plan modify my code while planning?

No, the skill is planning-only by contract. The only repository file it writes is the plan document under docs/plans/, and the only other permitted state change is an index refresh via analyze --index-only.

How do I strengthen an existing plan with Deepen mode?

Run /gitnexus-plan deepen followed by the plan path. The skill re-reads the plan through its descriptor-anchored helper, re-verifies claims against a refreshed graph, reconciles landed execution steps, and rewrites the same file with a verified backup.

What are the limitations of PDG analysis in gitnexus-plan?

The pdg_query tool is intra-procedural, so cross-function flow comes from explain taint analysis or inter-procedural impact reach. If the PDG layer is missing after one refresh, the plan states so and skips statement-level claims.