Command line¶

neosloc PATH... assesses one or more repositories. Everything below is generated from neosloc --help when the documentation is built.

usage: neosloc [-h] [--json] [--schema] [-v] [--only ONLY] [--salary SALARY]
               [--overhead OVERHEAD] [--buy-price USD] [--horizon YEARS]
               [--make-model MODEL] [--version]
               [--log-level {debug,info,warning,error}] [-q]
               [--log-format {text,json}] [--agentic] [--judge] [--review]
               [--scope {docs,source,both}] [--tasks TASKS]
               [--probe-model PROBE_MODEL] [--judge-model JUDGE_MODEL]
               [--review-model REVIEW_MODEL]
               [--effort {low,medium,high,xhigh,max}] [--max-turns MAX_TURNS]
               [--max-cost MAX_COST] [--base-url BASE_URL] [--allow-writes]
               [--auth-header AUTH_HEADER] [--no-fallbacks]
               [--transcripts TRANSCRIPTS]
               [paths ...]

Integrability evaluation for software repositories.

positional arguments:
  paths                 repository directories to assess

options:
  -h, --help            show this help message and exit

output:
  --json                emit JSON (reports and errors) instead of text
  --schema              print the JSON Schema of --json output and exit
  -v, --verbose         show all evidence in the text report
  --only ONLY           comma-separated dimension keys to run (interface,
                        stability, events, identity, portability, ergonomics,
                        embeddability, extensibility, observability,
                        legibility)
  --salary SALARY       annual salary for cost figures (default: sloccount's
                        56286)
  --overhead OVERHEAD   overhead multiplier on salary (default: 2.4)
  --buy-price USD       price per year of buying/adopting the product, for
                        make-or-buy (default: 0)
  --horizon YEARS       make-or-buy horizon in years (default: 3.0)
  --make-model MODEL    model whose prices cost the 'make' tokens (default:
                        anthropic:claude-opus-5-5)
  --version             show program's version number and exit

diagnostics (stderr):
  --log-level {debug,info,warning,error}
                        debug: detector timings; info (default): evaluator
                        progress; warning; error. Also NEOSLOC_LOG_LEVEL.
  -q, --quiet           only errors on stderr (same as --log-level error)
  --log-format {text,json}
                        text (default) or one JSON object per line

LLM evaluators (call model APIs and cost money):
  Models are written provider:model, e.g. anthropic:claude-opus-5-5 or
  openrouter:google/gemini-3.8-flash; a comma-separated list runs a panel.
  Anthropic needs ANTHROPIC_API_KEY and the [agentic] extra; OpenRouter
  needs OPENROUTER_API_KEY.

  --agentic             probe: have a model attempt the standard integration
                        tasks and grade the answers
  --judge               judge: have a model check each grounded probe answer
                        against the source (implies --agentic)
  --review              review: have a model audit each static dimension level
                        against the code
  --scope {docs,source,both}
                        what the agent may read (default: docs; 'both'
                        measures the documentation gap)
  --tasks TASKS         comma-separated task keys (default: all)
  --probe-model PROBE_MODEL, --model PROBE_MODEL
                        probe model(s) (default: anthropic:claude-opus-5-5)
  --judge-model JUDGE_MODEL
                        judge model (default: the first probe model)
  --review-model REVIEW_MODEL
                        review model(s) (default: anthropic:claude-opus-5-5)
  --effort {low,medium,high,xhigh,max}
                        effort level (default: medium)
  --max-turns MAX_TURNS
                        per task (default: 25)
  --max-cost MAX_COST   shared budget for all model calls in USD; no new task
                        or review starts once it is spent (default: 5.0)
  --base-url BASE_URL   probe a running instance over HTTP
  --allow-writes        permit POST/PUT/PATCH/DELETE against --base-url
  --auth-header AUTH_HEADER
                        'Name: value' header added to every live request
                        (repeatable)
  --no-fallbacks        don't let the API retry refused requests on a fallback
                        model
  --transcripts TRANSCRIPTS
                        directory to save per-task transcripts

Exit status: 0 ok, 1 internal error, 2 usage error, 3 a path could not be
assessed, 4 LLM evaluator setup error (credentials, provider, model, SDK), 130
interrupted. With --json, errors are JSON on stdout too; `neosloc --schema`
describes every shape.

Output and diagnostics¶

stdout carries the result: the text report, or with --json one JSON document, including errors (JSON output).
stderr carries diagnostics through the neosloc logger.

Level	What you see
`error`	Fatal errors. `-q`/`--quiet` shows only these.
`warning`	Degraded results, such as a budget exhausted before every task ran.
`info` (default)	Progress of LLM evaluators. Static runs print nothing here.
`debug`	Per-detector levels and timings, inventory size, total time, tracebacks of internal errors.

Set the level with --log-level or NEOSLOC_LOG_LEVEL; --quiet wins over both. --log-format json writes one JSON object per line (ts, level, logger, message, plus fields such as detector, seconds, code), which is useful in CI logs.

Exit status¶

Status	Meaning	JSON `error.code`
`0`	Every path was assessed.
`1`	Internal error, a bug in neosloc.	`internal`
`2`	Invalid command line.	`usage`, `unknown_dimension`, `unknown_task`, `bad_model_spec`
`3`	At least one path couldn't be assessed; the others are still reported.	`not_a_directory`
`4`	LLM evaluator setup failed.	`missing_dependency`, `credentials`, `provider_unreachable`, `unknown_model`, `model_unsupported`
`130`	Interrupted.	`interrupted`

Examples¶

neosloc .                                   # text report for the current repository
neosloc --json ~/src/a ~/src/b > report.json
neosloc --schema > neosloc.schema.json      # the shape of every --json document
neosloc -q --json . | jq '.dimensions[] | {key, level}'
neosloc --log-level debug --log-format json . 2> timings.jsonl
neosloc -v --only interface,stability .     # all evidence, two dimensions
neosloc --salary 90000 --overhead 1.8 .     # value in your own cost terms
neosloc --agentic --scope both --transcripts runs/ .
neosloc --review --review-model openrouter:google/gemini-3.8-flash,openrouter:openai/gpt-5.6-luna .