trueline-mcp 2.4.4 → 2.5.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -37,7 +37,7 @@ matches) the agent silently corrupts your code.
37
37
 
38
38
  ## How trueline fixes this
39
39
 
40
- trueline replaces the built-in `Read` and `Edit` with five tools that
40
+ trueline replaces the built-in `Read` and `Edit` with six tools that
41
41
  are smaller, faster, and verified.
42
42
 
43
43
  ### Read less: `trueline_outline` + `trueline_read`
@@ -100,6 +100,37 @@ disk.
100
100
 
101
101
  No more silent corruption. No more ambiguous string matches.
102
102
 
103
+ `trueline_verify` lets the agent check whether held checksums are still
104
+ valid without re-reading the file. When the file hasn't changed (the
105
+ common case), the response is a single line — near-zero tokens.
106
+
107
+ ### Benchmarks
108
+
109
+ Measured on real project files (`src/streaming-edit.ts`, 529 lines),
110
+ comparing total bytes through the context window (call + result, ÷4 ≈
111
+ tokens):
112
+
113
+ | Workflow | Built-in | Trueline | Saved |
114
+ |-------------------------|----------|-----------|-------|
115
+ | Navigate & understand | 22 094 | 3 609 | 84% |
116
+ | Explore then edit | 22 729 | 8 515 | 63% |
117
+ | Search & fix | 22 731 | 812 | 96% |
118
+ | Multi-region read | 22 094 | 2 720 | 88% |
119
+ | Multi-file exploration | 39 296 | 1 761 | 96% |
120
+ | Verify before edit | 44 823 | 3 608 | 92% |
121
+ | **Total** |**173 767**| **21 025**| **88%** |
122
+
123
+ The search-and-fix workflow saves the most: a single `trueline_search`
124
+ call replaces grep + full-file read + old-string echo, cutting **96%**
125
+ of token usage. The verify-before-edit workflow shows how
126
+ `trueline_verify` avoids a full re-read when checking whether held
127
+ checksums are still valid — **92%** savings over re-reading the entire
128
+ file. Even the explore-then-edit workflow — which includes an
129
+ exploratory read, a targeted re-read, and an edit — still saves **63%**
130
+ over the built-in equivalent.
131
+
132
+ Run the benchmark yourself: `bun run benchmarks/token-benchmark.ts`
133
+
103
134
  ## Design
104
135
 
105
136
  See [DESIGN.md](DESIGN.md) for the protocol specification, hash