sourcecode 1.8.0

Deterministic codebase context for AI coding agents

sourcecode

Deterministic codebase context for AI coding agents.

---

What is it?

sourcecode analyzes a repository and produces a structured context map (JSON or YAML) designed to be consumed by AI agents or language models. It solves the "stuff the whole repo into the prompt" problem by instead producing a deterministic extract: entry points, dependencies, stacks, inline annotations, environment variables, and git activity. It is an MVP tool under active evolution — the semantic analysis and module graph features work but have known limitations that are explicitly documented below.

---

Installation

Prerequisites: Python 3.10+

pip install sourcecode
# or with pipx for isolation:
pipx install sourcecode

Verify installation:

sourcecode version
# sourcecode 1.0.0

---

Quickstart

The most useful command for integrating sourcecode into an AI agent:

sourcecode --agent

It produces a structured JSON with the essential sections (no noise, no file tree), ready to paste into an LLM context:

{
  "project": {
    "type": "fullstack",
    "summary": "Full-stack project in Nodejs, mvc, 4075 source files. Domains: atlas-client, atlas-server, atlas-hub, atlas-reports. 3300 dependencies (java, nodejs).",
    "primary_stack": "nodejs",
    "secondary_stacks": ["java"]
  },
  "entry_points": [
    {
      "path": "atlas-server/src/main/java/com/example/atlas/AtlasServerApplication.java",
      "stack": "java",
      "kind": "application",
      "confidence": "high"
    },
    {
      "path": "atlas-client/src/main.ts",
      "stack": "nodejs",
      "kind": "entrypoint",
      "confidence": "high"
    }
  ],
  "runtime_packages": [ ... ],
  "dependencies": { ... },
  "env_map": { ... },
  "code_notes": [ ... ]
}

For large repositories where context matters, use --compact to reduce to ~600-800 tokens:

sourcecode --compact --copy
# Copies the summary to the clipboard. Ready to paste.

---

Flags reference

Global options

Flag	Alias	Type	Default	Description	Status
--format	-f	json|yaml	json	Output format. YAML is more readable, JSON preferred in pipelines.	✅ CORE
--output	-o	PATH	stdout	Writes output to a file instead of stdout.	✅ CORE
--compact		flag	off	~600-800 token output: stacks, entry points, deps, gaps. No file tree.	✅ CORE
--agent		flag	off	JSON optimized for agents. Automatically enables --dependencies, --env-map, --code-notes.	✅ CORE
--dependencies		flag	off	Analyzes direct and transitive deps from manifests and lockfiles.	✅ CORE
--git-context	-g	flag	off	Includes recent commits, change hotspots, uncommitted changes, contributors.	✅ CORE
--git-depth		INT [1–100]	20	Number of recent commits with --git-context.	✅ CORE
--git-days		INT [1–3650]	90	Window in days to detect hotspots with --git-context.	✅ CORE
--env-map		flag	off	Maps environment variables: key, type, category, files that reference them.	✅ CORE
--code-notes		flag	off	Extracts inline annotations: TODO, FIXME, HACK, BUG, DEPRECATED, NOTE, etc.	✅ CORE
--copy	-c	flag	off	Copies output to the clipboard after successful execution.	✅ CORE
--depth		INT [1–20]	4	Maximum file tree traversal depth. Java/Maven requires ≥8.	✅ CORE
--mode		contract|standard|raw	contract	contract: minimal contracts per file. standard: full detail. raw: project level only.	✅ CORE
--tree		flag	off	Includes full file_tree and file_paths in the output. Increases size significantly.	✅ CORE
--changed-only		flag	off	Contract mode: only files modified in git (staged, unstaged, untracked).	✅ CORE
--rank-by		relevance|centrality|git-churn	relevance	File ranking strategy in contract mode.	✅ CORE
--semantics		flag	off	Cross-file symbol resolution, call graph with confidence levels, fan-in/fan-out hotspots. Slower.	🧪 EXP
--architecture		flag	off	Architectural layer inference (MVC/hexagonal/bounded contexts). Low confidence without --semantics.	🧪 EXP
--graph-modules		flag	off	Structural module graph: nodes (files/symbols) and edges (imports, calls, contains).	🧪 EXP
--graph-detail		high|medium|full	high	Module graph detail level.	🧪 EXP
--max-nodes		INT [≥1]	—	Maximum nodes in --graph-modules. Prevents huge graphs in large repos.	🧪 EXP
--graph-edges		TEXT	all	Edge types for --graph-modules, comma-separated: imports,calls,contains.	🧪 EXP
--docs		flag	off	Extracts docstrings, function signatures, and module comments.	🧪 EXP
--docs-depth		module|symbols|full	symbols	Docs extraction depth. full includes private symbols.	🧪 EXP
--symbol		TEXT	—	Contract mode: localized context for a specific symbol. Python, TS, JS only. Does not support Java.	🧪 EXP
--max-importers		INT [1–10000]	50	Limit on importer files returned by --symbol.	🧪 EXP
--full-metrics		flag	off	Per-file technical metrics: LOC, cyclomatic complexity, coverage. Aimed at CI, not at agents.	🧪 EXP
--emit-graph		flag	off	Contract mode: includes a compact dependency graph (nodes + edges) in the output.	🚧 WIP
--entrypoints-only		flag	off	Contract mode: only files with exports or entry points. Note: includes all files with exports.	🚧 WIP
--max-symbols		INT [≥1]	—	Limits total exported symbols in contract mode. Discards lower-ranked files.	🚧 WIP
--no-redact		flag	off	Disables automatic secret redaction. Output may contain sensitive values.	🚧 WIP
--trace-pipeline		flag	off	Diagnostic mode: includes a trace of each candidate and filtering decision. Debugging only.	🚧 WIP
--version	-v	flag	—	Shows version and exits.	✅ CORE

---

Subcommands

prepare-context TASK [PATH]

Generates task-specific context for AI agents.

Task	Description	Status
explain	Architecture, entry points, key dependencies	✅ CORE
fix-bug	Files prioritized by risk, inline annotations	✅ CORE
onboard	Full context for new agents or developers	✅ CORE
delta	Incremental context: only files changed in git	✅ CORE
refactor	Structural problems, improvement opportunities	🧪 EXP
generate-tests	Files without tests, coverage gap analysis	🧪 EXP
review-pr	Changed files + architectural impact	🧪 EXP

...