tree-sitter-analyzer: Skill for Claude Code

tsa-landing — Land in any repo in <3s

First action on entering a new repo. Replaces 6 separate bootstrap calls (~15k tokens) with 3-4 parallel calls (~2k tokens). 92% token saved.

When to use

• You just entered an unfamiliar project (no Claude.md / no skim yet)
• You return after long gap and need "what's the state?"
• User asks any of: "what is this project / where do I start / what's the entry / recent changes / health"

Don't use when:

• You already have full context (just continue working)
• User wants to read source — use structure action=read directly

Procedure

Use the MCP server's configured project root. TSA's native discovery does not require fd or rg; do not run an external-search dependency preflight or stop because those executables are absent. The first project overview call below establishes whether the configured project can be read. Indexed workflow map steps build or update the AST index before requesting a sitemap.

Step 1 — Fan-out 4 MCP calls (in single message, parallel)

Call these 4 tools in ONE message (parallel tool use):

1. project action=overview (no args) — project_card + entry_points
2. health action=project with max_files: 5 — grade distribution + weakest dimension
3. edit action=impact with mode: "branch" — recent_signals (last commit, ahead-of-main)
4. project action=workflow (no args) — current_phase + recommended_commands

Step 2 — Fold and emit decision_surface

Combine into single Decision Surface:

{
  "project_card": {
    "name": <from project action=overview project_root basename>,
    "primary_language": <key with the highest count in project action=overview summary.by_language — it is a {language: file_count} dict sorted descending, so the first key>,
    "language_mix": <from project action=overview summary.by_language — first 3 keys of the {language: file_count} dict>,
    "size": {
      "files": <from project action=overview summary.total_files>,
      "loc": <from project action=overview summary.total_lines>
    }
  },
  "entry_points": <from project action=overview entry_points>,
  "recent_signals": {
    "last_commit": <git log -1 --oneline via Bash>,
    "ahead_of_origin": <git rev-list --count via Bash>,
    "uncommitted_files": <git status --short | wc -l via Bash>,
    "branch": <git branch --show-current via Bash>
  },
  "health": {
    "verdict": <from health action=project verdict>,
    "risk": <from health action=project agent_summary.risk>,
    "grade_distribution": <from health action=project grade_distribution>,
    "weakest_dimension": <from health action=project weakest_dimension>
  },
  "top_files_to_know": [
    "AGENTS.md",
    "CLAUDE.md",
    <from project action=overview entry_points>,
    <top 3 from health action=project top_refactoring_targets>
  ],
  "agent_next_step": {
    "if_asked_what_is_this":
      "Read AGENTS.md (canonical contracts) + docs/CODEMAPS/architecture.md (topology). Stop after 2k tokens.",
    "if_asked_to_add_feature":
      "Call project action=workflow → follow phase_order. Use TDD (write test first).",
    "if_asked_to_fix_bug":
      "Call health action=patterns file_path=<file> → edit action=refactor file_path=<file>. Cross-ref nav action=lineage if symbol-level.",
    "if_asked_about_test_status":
      "Run: uv run pytest -q (5-min cap). Project enforces xdist parallel, ~5min for 15k tests."
  },
  "summary_line": "<project> files=<N> py=<X%> grade=<G> recent=<commit_subj>",
  "verdict": "INFO"
}