refactor: remove all caveman references from codebase

CHANGELOG.md

@@ -16,13 +16,13 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Dollar-based cost savings tracking in Stats
 - Model-aware BPE encoding selection
 
-### Added — Round 2 of rtk + caveman porting (2026-06-01)
-- **`tok.CompressCaveman(text, intensity)`** — public Go API for the
-  caveman prompt-compression algorithm. Three intensity levels
-  (`CavemanLite`, `CavemanFull`, `CavemanUltra`) with drop-lists for
+### Added — Round 2 porting (2026-06-01)
+- **`tok.PromptCompress(text, intensity)`** — public Go API for the
+  prompt-compression prompt-compression algorithm. Three intensity levels
+  (`IntensityLite`, `IntensityFull`, `IntensityUltra`) with drop-lists for
   articles / filler / pleasantries and ~150 phrase substitutions.
   Auto-clarity: security / destructive segments pass through verbatim.
-  Returns a `CavemanStats` struct (OriginalBytes, CompressedBytes,
+  Returns a `PromptStats` struct (OriginalBytes, CompressedBytes,
   BytesSaved, PercentOff, PassThroughSegments, etc.).
 - **`tok.IsSensitiveFilename(path)`** — 3-layer path-based sensitive
   detection (exact basename, sensitive directory, name token).
@@ -83,7 +83,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 * add agent/model/provider attribution for token savings ([c8213f1](https://github.com/GrayCodeAI/tok/commit/c8213f1d84f8c8e0378852ae4a701588f4fb294b))
 * add ALL 152 official git commands ([b48bc18](https://github.com/GrayCodeAI/tok/commit/b48bc189e54976a5280d389cc8365fc866055744))
 * add all competitor features - tee recovery, cost tracking, live monitor, prompt debugger, discover, session tracking, dup detection ([dc23fda](https://github.com/GrayCodeAI/tok/commit/dc23fda6f237ebb2b529386bad7ceb3d64da3a0c))
-* add all rtk + caveman features to tok ([87b26bf](https://github.com/GrayCodeAI/tok/commit/87b26bf8b9dfcb4419e29cda1edd3758212b95d6))
+* add all rtk + prompt-compression features to tok ([87b26bf](https://github.com/GrayCodeAI/tok/commit/87b26bf8b9dfcb4419e29cda1edd3758212b95d6))
 * add all RTK features and performance optimizations ([98295cd](https://github.com/GrayCodeAI/tok/commit/98295cd93d2f341b4c34b27e6e3b5919b21d9d4f))
 * add beautiful real-time TUI dashboard ([01bd0d8](https://github.com/GrayCodeAI/tok/commit/01bd0d84b9de5786a486e9404f5d3d869e7844b5))
 * add benchmark CI reporting and adaptive profile validation ([bb1ca2f](https://github.com/GrayCodeAI/tok/commit/bb1ca2f850343b5f591491e0f36642fca8056198))
@@ -158,9 +158,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 * Claude Code-style interface with open input box ([e46fde2](https://github.com/GrayCodeAI/tok/commit/e46fde2cdcacbba9d842e05899755d8e7b94dd98))
 * clean chat-style CLI like Mistral Vibe ([fc2228e](https://github.com/GrayCodeAI/tok/commit/fc2228e32dff4a2b4c4cc4bea49ff5b489d3c8bb))
 * **cli:** add compaction flags for Layer 11 ([f9c868c](https://github.com/GrayCodeAI/tok/commit/f9c868c987fe478109628ef9ba6baeb689f576c2))
-* close 9 world-class gaps vs rtk and caveman ([f0059eb](https://github.com/GrayCodeAI/tok/commit/f0059ebc2201cbe3c0ad5a8776b5f656e1bc8c87))
-* close last two caveman gaps ([3028726](https://github.com/GrayCodeAI/tok/commit/302872695f5482bb8a8e5c340e1613755a31dd1c))
-* close tok gaps vs rtk and caveman ([760f326](https://github.com/GrayCodeAI/tok/commit/760f3268c5de816c00e6f92657f8f19834457942))
+* close 9 world-class gaps vs rtk and prompt-compression ([f0059eb](https://github.com/GrayCodeAI/tok/commit/f0059ebc2201cbe3c0ad5a8776b5f656e1bc8c87))
+* close last two prompt-compression gaps ([3028726](https://github.com/GrayCodeAI/tok/commit/302872695f5482bb8a8e5c340e1613755a31dd1c))
+* close tok gaps vs rtk and prompt-compression ([760f326](https://github.com/GrayCodeAI/tok/commit/760f3268c5de816c00e6f92657f8f19834457942))
 * complete honest competitor analysis - cloned and analyzed 15 OSS tools ([3000897](https://github.com/GrayCodeAI/tok/commit/3000897d43294cebc80fe4ad37f8c22142e2e642))
 * complete integration test suite with archive fixes and MCP tools ([669293c](https://github.com/GrayCodeAI/tok/commit/669293c6fe9589696197e2dcbe9f22c5d5b71535))
 * complete SIMD optimization with Go 1.26 ([a3dbc85](https://github.com/GrayCodeAI/tok/commit/a3dbc857d662ab98690cc06e90d888491bf0f0fa))
@@ -474,7 +474,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 * add agent/model/provider attribution for token savings ([c8213f1](https://github.com/GrayCodeAI/tok/commit/c8213f1d84f8c8e0378852ae4a701588f4fb294b))
 * add ALL 152 official git commands ([b48bc18](https://github.com/GrayCodeAI/tok/commit/b48bc189e54976a5280d389cc8365fc866055744))
 * add all competitor features - tee recovery, cost tracking, live monitor, prompt debugger, discover, session tracking, dup detection ([dc23fda](https://github.com/GrayCodeAI/tok/commit/dc23fda6f237ebb2b529386bad7ceb3d64da3a0c))
-* add all rtk + caveman features to tok ([87b26bf](https://github.com/GrayCodeAI/tok/commit/87b26bf8b9dfcb4419e29cda1edd3758212b95d6))
+* add all rtk + prompt-compression features to tok ([87b26bf](https://github.com/GrayCodeAI/tok/commit/87b26bf8b9dfcb4419e29cda1edd3758212b95d6))
 * add all RTK features and performance optimizations ([98295cd](https://github.com/GrayCodeAI/tok/commit/98295cd93d2f341b4c34b27e6e3b5919b21d9d4f))
 * add beautiful real-time TUI dashboard ([01bd0d8](https://github.com/GrayCodeAI/tok/commit/01bd0d84b9de5786a486e9404f5d3d869e7844b5))
 * add benchmark CI reporting and adaptive profile validation ([bb1ca2f](https://github.com/GrayCodeAI/tok/commit/bb1ca2f850343b5f591491e0f36642fca8056198))
@@ -548,9 +548,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 * Claude Code-style interface with open input box ([e46fde2](https://github.com/GrayCodeAI/tok/commit/e46fde2cdcacbba9d842e05899755d8e7b94dd98))
 * clean chat-style CLI like Mistral Vibe ([fc2228e](https://github.com/GrayCodeAI/tok/commit/fc2228e32dff4a2b4c4cc4bea49ff5b489d3c8bb))
 * **cli:** add compaction flags for Layer 11 ([f9c868c](https://github.com/GrayCodeAI/tok/commit/f9c868c987fe478109628ef9ba6baeb689f576c2))
-* close 9 world-class gaps vs rtk and caveman ([f0059eb](https://github.com/GrayCodeAI/tok/commit/f0059ebc2201cbe3c0ad5a8776b5f656e1bc8c87))
-* close last two caveman gaps ([3028726](https://github.com/GrayCodeAI/tok/commit/302872695f5482bb8a8e5c340e1613755a31dd1c))
-* close tok gaps vs rtk and caveman ([760f326](https://github.com/GrayCodeAI/tok/commit/760f3268c5de816c00e6f92657f8f19834457942))
+* close 9 world-class gaps vs rtk and prompt-compression ([f0059eb](https://github.com/GrayCodeAI/tok/commit/f0059ebc2201cbe3c0ad5a8776b5f656e1bc8c87))
+* close last two prompt-compression gaps ([3028726](https://github.com/GrayCodeAI/tok/commit/302872695f5482bb8a8e5c340e1613755a31dd1c))
+* close tok gaps vs rtk and prompt-compression ([760f326](https://github.com/GrayCodeAI/tok/commit/760f3268c5de816c00e6f92657f8f19834457942))
 * complete honest competitor analysis - cloned and analyzed 15 OSS tools ([3000897](https://github.com/GrayCodeAI/tok/commit/3000897d43294cebc80fe4ad37f8c22142e2e642))
 * complete integration test suite with archive fixes and MCP tools ([669293c](https://github.com/GrayCodeAI/tok/commit/669293c6fe9589696197e2dcbe9f22c5d5b71535))
 * complete SIMD optimization with Go 1.26 ([a3dbc85](https://github.com/GrayCodeAI/tok/commit/a3dbc857d662ab98690cc06e90d888491bf0f0fa))

caveman.go → compress.go

@@ -1,14 +1,14 @@
-// Caveman prompt compression — public API.
+// Prompt compression — public API.
 //
-// This file exposes the caveman algorithm (port from github.com/JuliusBrussee/caveman)
+// This file exposes the algorithm (port from prompt-compression algorithm)
 // as a top-level tok API. It is independent of the main compression pipeline
 // and does not use tiers/modes/budgets; it is a self-contained, deterministic
 // prose compression function with three intensity levels.
 //
 // Usage:
 //
-//	out, stats := tok.CompressCaveman("Sure, I can help you with that.", tok.CavemanLite)
-//	out, stats := tok.CompressCaveman(prompt, tok.CavemanUltra)
+//	out, stats := tok.PromptCompress("Sure, I can help you with that.", tok.IntensityLite)
+//	out, stats := tok.PromptCompress(prompt, tok.IntensityUltra)
 //
 // The function NEVER modifies sensitive segments (security warnings, destructive
 // commands, credential mentions) — those are passed through verbatim regardless
@@ -18,7 +18,7 @@ package tok
 
 import "github.com/GrayCodeAI/tok/internal/compress"
 
-// Intensity selects how aggressive caveman compression should be.
+// Intensity selects how aggressive prompt compression should be.
 //
 //   - IntensityLite:   drops pleasantries only ("Sure", "Of course", "Please").
 //   - IntensityFull:   Lite + articles ("a", "an", "the") + filler + hedging.
@@ -28,17 +28,17 @@ import "github.com/GrayCodeAI/tok/internal/compress"
 // a longer output than a lower intensity on the same input.
 type Intensity = compress.Intensity
 
-// Intensity preset constants. Use as a direct argument to CompressCaveman.
+// Intensity preset constants. Use as a direct argument to PromptCompress.
 const (
-	CavemanLite  Intensity = compress.Lite
-	CavemanFull  Intensity = compress.Full
-	CavemanUltra Intensity = compress.Ultra
+	IntensityLite  Intensity = compress.Lite
+	IntensityFull  Intensity = compress.Full
+	IntensityUltra Intensity = compress.Ultra
 )
 
-// CavemanStats is the statistics struct returned from CompressCaveman.
-type CavemanStats = compress.Stats
+// PromptStats is the statistics struct returned from PromptCompress.
+type PromptStats = compress.Stats
 
-// CompressCaveman applies the caveman prompt-compression algorithm to text.
+// PromptCompress applies the prompt-compression algorithm to text.
 //
 // Behavior:
 //   - Empty input returns ("", zero stats).
@@ -48,6 +48,6 @@ type CavemanStats = compress.Stats
 //   - Dictionary substitutions preserve the case of the first character.
 //
 // This is independent of tok.Compress() and does not consume options/presets.
-func CompressCaveman(text string, intensity Intensity) (string, CavemanStats) {
+func PromptCompress(text string, intensity Intensity) (string, PromptStats) {
 	return compress.Compress(text, intensity)
 }

caveman_test.go → compress_test.go

@@ -7,32 +7,32 @@ import (
 	"github.com/GrayCodeAI/tok"
 )
 
-func TestCompressCaveman_Lite(t *testing.T) {
+func TestPromptCompress_Lite(t *testing.T) {
 	in := "Sure, I can help you with that. Of course, the answer is yes."
-	out, stats := tok.CompressCaveman(in, tok.CavemanLite)
+	out, stats := tok.PromptCompress(in, tok.IntensityLite)
 	if strings.Contains(out, "Sure,") {
 		t.Errorf("expected 'Sure,' to be dropped at Lite, got %q", out)
 	}
 	if !strings.Contains(out, "help you with that") {
 		t.Errorf("expected help text preserved, got %q", out)
 	}
-	if stats.Intensity != tok.CavemanLite {
+	if stats.Intensity != tok.IntensityLite {
 		t.Errorf("expected intensity=Lite, got %v", stats.Intensity)
 	}
 }
 
-func TestCompressCaveman_Full(t *testing.T) {
+func TestPromptCompress_Full(t *testing.T) {
 	in := "The quick brown fox jumps over the lazy dog."
-	out, _ := tok.CompressCaveman(in, tok.CavemanFull)
+	out, _ := tok.PromptCompress(in, tok.IntensityFull)
 	// "the" should be dropped
 	if strings.Contains(out, " the ") && !strings.Contains(out, "over") {
 		t.Errorf("expected 'the' to be dropped at Full, got %q", out)
 	}
 }
 
-func TestCompressCaveman_Ultra(t *testing.T) {
+func TestPromptCompress_Ultra(t *testing.T) {
 	in := "However, the system is basically quite slow. Therefore, we need to optimize it."
-	out, stats := tok.CompressCaveman(in, tok.CavemanUltra)
+	out, stats := tok.PromptCompress(in, tok.IntensityUltra)
 	if strings.Contains(out, "However,") {
 		t.Errorf("expected 'However,' to be dropped at Ultra, got %q", out)
 	}
@@ -44,9 +44,9 @@ func TestCompressCaveman_Ultra(t *testing.T) {
 	}
 }
 
-func TestCompressCaveman_SensitivePassThrough(t *testing.T) {
+func TestPromptCompress_SensitivePassThrough(t *testing.T) {
 	in := "Be careful with rm -rf /tmp. The file is large."
-	out, stats := tok.CompressCaveman(in, tok.CavemanFull)
+	out, stats := tok.PromptCompress(in, tok.IntensityFull)
 	// The segment containing "rm -rf" must be preserved verbatim.
 	if !strings.Contains(out, "rm -rf /tmp.") {
 		t.Errorf("expected 'rm -rf /tmp.' to be preserved, got %q", out)
@@ -56,8 +56,8 @@ func TestCompressCaveman_SensitivePassThrough(t *testing.T) {
 	}
 }
 
-func TestCompressCaveman_Empty(t *testing.T) {
-	out, stats := tok.CompressCaveman("", tok.CavemanFull)
+func TestPromptCompress_Empty(t *testing.T) {
+	out, stats := tok.PromptCompress("", tok.IntensityFull)
 	if out != "" {
 		t.Errorf("expected empty output, got %q", out)
 	}
@@ -66,9 +66,9 @@ func TestCompressCaveman_Empty(t *testing.T) {
 	}
 }
 
-func TestCompressCaveman_Dictionary(t *testing.T) {
+func TestPromptCompress_Dictionary(t *testing.T) {
 	in := "In order to install, you need to make use of the installer."
-	out, _ := tok.CompressCaveman(in, tok.CavemanFull)
+	out, _ := tok.PromptCompress(in, tok.IntensityFull)
 	if strings.Contains(out, "In order to") {
 		t.Errorf("expected 'In order to' to be replaced, got %q", out)
 	}
@@ -77,21 +77,21 @@ func TestCompressCaveman_Dictionary(t *testing.T) {
 	}
 }
 
-func TestCompressCaveman_DoesNotAffectTopLevelCompress(t *testing.T) {
-	// Regression: adding caveman API must not change the main Compress() output
+func TestPromptCompress_DoesNotAffectTopLevelCompress(t *testing.T) {
+	// Regression: adding prompt-compression API must not change the main Compress() output
 	// for a typical input. We test that both APIs can be called and return
-	// different shapes (caveman returns CavemanStats, main returns Stats).
+	// different shapes (PromptCompress returns PromptStats, main returns Stats).
 	in := "The quick brown fox jumps over the lazy dog."
 	mainOut, mainStats := tok.Compress(in, tok.Aggressive)
 	if mainOut == "" {
 		t.Error("expected non-empty main Compress output")
 	}
 	_ = mainStats.OriginalTokens
 
-	cavemanOut, _ := tok.CompressCaveman(in, tok.CavemanFull)
-	// Caveman output may equal main output (if both drop "The") or differ —
+	promptCompressOut, _ := tok.PromptCompress(in, tok.IntensityFull)
+	// PromptCompress output may equal main output (if both drop "The") or differ —
 	// the point is they don't crash each other.
-	if cavemanOut == "" {
-		t.Error("expected non-empty caveman output")
+	if promptCompressOut == "" {
+		t.Error("expected non-empty prompt-compression output")
 	}
 }

internal/compress/caveman.go → internal/compress/algorithm.go

@@ -1,5 +1,5 @@
-// Package compress implements the caveman prompt-compression algorithm
-// ported natively to Go. Source: github.com/JuliusBrussee/caveman (MIT).
+// Package compress implements the prompt-compression algorithm
+// ported natively to Go. Source: JuliusBrussee/prompt-compression (MIT).
 //
 // Algorithm overview:
 //
@@ -17,7 +17,7 @@ import (
 	"strings"
 )
 
-// CompressResult reports what caveman did to the input.
+// CompressResult reports what the compression did to the input.
 type CompressResult struct {
 	// Original is the input verbatim.
 	Original string
@@ -50,11 +50,11 @@ type Stats struct {
 	SensitiveKeywordsHit []string
 }
 
-// Compress applies the caveman algorithm to s at the given intensity.
+// Compress applies the algorithm to s at the given intensity.
 //
 // Behavior:
 //   - Empty input returns (input, empty result) untouched.
-//   - CJK-heavy text is passed through (caveman rules assume Latin grammar).
+//   - CJK-heavy text is passed through (compression rules assume Latin grammar).
 //   - Sensitive segments (security/destructive keywords) are preserved
 //     verbatim regardless of intensity.
 //   - Multi-word drop-list entries ("of course", "feel free") are matched

internal/compress/caveman_dict.go → internal/compress/dict.go

@@ -1,7 +1,7 @@
 package compress
 
-// engV1Dictionary is the caveman "eng/v1" phrase-substitution dictionary.
-// Source: caveman/skills/caveman/SKILL.md, "Dictionary" section.
+// engV1Dictionary is the "eng/v1" phrase-substitution dictionary.
+// Source: skills/prompt-compression/SKILL.md, "Dictionary" section.
 //
 // Keys are case-insensitive verbose phrases. Values are terse equivalents.
 // Multi-word keys are matched as whole phrases (whitespace-normalized).

internal/compress/caveman_rules.go → internal/compress/rules.go

@@ -1,5 +1,5 @@
-// Package compress implements the caveman prompt-compression algorithm
-// ported natively to Go. Source: github.com/JuliusBrussee/caveman (MIT).
+// Package compress implements the prompt-compression algorithm
+// ported natively to Go. Source: JuliusBrussee/prompt-compression (MIT).
 //
 // The algorithm is a pure text-rewrite engine. Three intensity levels
 // (Lite, Full, Ultra) progressively drop articles, filler words,
@@ -10,7 +10,7 @@
 // warnings, destructive operations, and code-bearing segments.
 package compress
 
-// Intensity controls the aggressiveness of caveman compression.
+// Intensity controls the aggressiveness of prompt compression.
 //
 // The intensity ordering is Lite < Full < Ultra. Higher intensity drops
 // more word classes and applies more aggressive dictionary substitutions.
@@ -20,7 +20,7 @@ const (
 	// Lite drops only pleasantries and obvious filler; preserves articles.
 	Lite Intensity = iota
 	// Full drops articles, filler, pleasantries, and hedging.
-	// This is the caveman default.
+	// This is the default.
 	Full
 	// Ultra additionally drops conjunctions and forces sentence fragments.
 	Ultra
@@ -41,7 +41,7 @@ func (i Intensity) String() string {
 }
 
 // dropLists maps each intensity to the set of words/phrases to drop.
-// Source: caveman/skills/caveman/SKILL.md, "Intensity" section.
+// Source: skills/prompt-compression/SKILL.md, "Intensity" section.
 //
 // Multi-word phrases ("of course", "feel free") are matched case-insensitive
 // as whole phrases, not as individual words.

internal/compress/caveman_safety.go → internal/compress/safety.go

@@ -8,7 +8,7 @@ import (
 // safetyKeywords are substrings that, if present, force the compressor
 // to fall back to "normal" prose (no compression) for that segment.
 //
-// Source: caveman/skills/caveman/SKILL.md, "Auto-clarity rules":
+// Source: skills/prompt-compression/SKILL.md, "Auto-clarity rules":
 //
 //	"Drop to normal prose for security warnings, destructive ops,
 //	ambiguous sequences."
@@ -116,7 +116,7 @@ func SplitSafeSegments(s string) []segment {
 	if s == "" {
 		return nil
 	}
-	// Split on sentence boundaries; the caveman rule operates per sentence.
+	// Split on sentence boundaries; the rule operates per sentence.
 	segs := splitSentences(s)
 	out := make([]segment, 0, len(segs))
 	for _, seg := range segs {
@@ -222,7 +222,7 @@ func isUpperLetter(b byte) bool {
 }
 
 // isCJK reports whether s is primarily CJK characters (Chinese/Japanese/Korean).
-// Used to decide if the caveman rules even apply — CJK text doesn't have
+// Used to decide if the compression rules even apply — CJK text doesn't have
 // articles/filler to drop.
 func isCJK(s string) bool {
 	cjk := 0