git-slop 0.16.4

A deterministic repository token-defragmenter for humans and AI agents.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83

# Configuration Recipes

All examples are partial overrides. Inspect defaults with `git slop config show --effective`.

## Small application

```yaml
schema_version: 2
organization:
  candidate_file_limit: 250
```

## Mature monorepo

```yaml
schema_version: 2
organization:
  max_commit_files: 150
  max_temporal_edges: 8000
resources:
  memory_budget_mb: 2048
output:
  retention_runs: 30
```

Use `git slop find --scope packages/example`; Git evidence remains repository-wide.

## Generated or vendor-heavy repository

```yaml
schema_version: 2
inventory:
  ignore_globs: [dist/**, vendor/**, "**/*.min.js"]
resources:
  large_file_bytes: 1048576
```

## Repository-specific path overrides

```yaml
schema_version: 2
inventory:
  path_overrides:
    - glob: "action/**/*.mjs"
      classification: source
      language: JavaScript
      verification_applicability: applicable
    - glob: "vendor/**"
      profile: data_context
      verification_applicability: not_applicable
```

Overrides are applied in order. The last matching value wins independently for classification, profile, language, and verification applicability.

## Documentation repository

```yaml
schema_version: 2
tokenization:
  context_bands:
    compact_max_tokens: 5000
    healthy_max_tokens: 12000
    warning_max_tokens: 18000
```

## Strict CI ratchet

Generate a compatible base report, pass it as the Action's `baseline-report`,
set `enforcement: regression`, and use `compare --fail-on-regression` locally.
This gates new or worsened findings without requiring the existing backlog to
disappear first. A healthy added file is not a regression.

## Fast local versus full-history

Use `--scope` and a smaller candidate limit for local feedback. Use a full clone, the normal candidate limit, and the same tokenizer/config digest for authoritative comparisons. Changing the tokenizer invalidates baselines because token counts and context bands are no longer comparable.

## Supported tokenizers and baseline compatibility

`tokenization.context_tokenizer_name` accepts `cl100k_base` (the default),
`o200k_base`, `o200k_harmony`, `p50k_base`, `p50k_edit`, or `r50k_base`.
The tokenizer name participates in analyzer provenance and the content-addressed
token cache. A tokenizer change deliberately invalidates cached counts and makes
reports comparison-incompatible unless `compare --force` is explicitly used.