Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Python agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
fix(crusher): switch compression_store cache key from MD5 to SHA-256 (CodeQL #395)
chopratejasaf9e6282f89f
diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.jsonindex 2b88649d8..1e91d756c 100644--- a/.claude-plugin/marketplace.json+++ b/.claude-plugin/marketplace.json@@ -5,14 +5,14 @@},"metadata": {"description": "Headroom marketplace for Claude Code and GitHub Copilot CLI plugins.",- "version": "0.20.16"+ "version": "0.20.28"},"plugins": [{"name": "headroom","source": "./plugins/headroom-agent-hooks","description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.",- "version": "0.20.16",+ "version": "0.20.28","author": {"name": "Headroom Contributors","url": "https://github.com/chopratejas/headroom"diff --git a/.github/plugin/marketplace.json b/.github/plugin/marketplace.jsonindex 2b88649d8..1e91d756c 100644--- a/.github/plugin/marketplace.json+++ b/.github/plugin/marketplace.json@@ -5,14 +5,14 @@},"metadata": {"description": "Headroom marketplace for Claude Code and GitHub Copilot CLI plugins.",- "version": "0.20.16"+ "version": "0.20.28"},"plugins": [{"name": "headroom","source": "./plugins/headroom-agent-hooks","description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.",- "version": "0.20.16",+ "version": "0.20.28","author": {"name": "Headroom Contributors","url": "https://github.com/chopratejas/headroom"diff --git a/headroom/cache/compression_store.py b/headroom/cache/compression_store.pyindex 4ab29ea70..810d1a07c 100644--- a/headroom/cache/compression_store.py+++ b/headroom/cache/compression_store.py@@ -199,7 +199,7 @@ class CompressionStore:compression_strategy: Strategy used for compression.ttl: Custom TTL in seconds (uses default if not specified).explicit_hash: Use this exact hex hash as the storage key- instead of computing MD5(original)[:24]. Required when+ instead of computing SHA-256(original)[:24]. Required whenthe marker that points at this entry was emitted by aproducer with its own hash function (e.g. SmartCrusher'sRust row-drop path uses SHA-256[:12]). If not a hex@@ -210,28 +210,35 @@ class CompressionStore:Returns:Hash key for retrieving this content."""- # Generate hash from original content. Default: MD5[:24] of the+ # Generate hash from original content. Default: SHA-256[:24] of the# original. When the caller provides `explicit_hash`, use it# verbatim — required when the hash that ends up in the prompt# marker is produced by another component (e.g. the Rust# SmartCrusher row-drop path emits SHA-256[:12], which the# Python store has to mirror so /v1/retrieve resolves it).- # CRITICAL FIX #5: Use 24 chars (96 bits) instead of 16 (64 bits) for better- # collision resistance. Birthday paradox: 50% collision at sqrt(2^n) entries.- # - 64 bits: ~4 billion entries for 50% collision- # - 96 bits: ~280 trillion entries for 50% collision+ # 24 chars (96 bits) was chosen for collision resistance under the+ # birthday bound: 50% collision probability at ~280 trillion entries+ # (2^48), versus ~4 billion (2^32) for the previous 16-char default.if explicit_hash is not None:# Validate as hex. Bail loudly per `feedback_no_silent_fallbacks`- # — silently falling back to MD5 when the caller asked for a- # specific key would defeat the marker/store consistency we're- # trying to preserve.+ # — silently falling back to the default hash when the caller+ # asked for a specific key would defeat the marker/store+ # consistency we're trying to preserve.if not explicit_hash or not all(c in "0123456789abcdefABCDEF" for c in explicit_hash):raise ValueError(f"explicit_hash must be a non-empty hex string, got {explicit_hash!r}")hash_key = explicit_hash.lower()else:- hash_key = hashlib.md5(original.encode()).hexdigest()[:24] # nosec B324+ # SHA-256 truncated to 24 hex chars (96 bits) — same collision+ # space as the MD5[:24] this replaced. Switched from MD5 in+ # PR #395 to silence CodeQL's `py/weak-sensitive-data-hashing`+ # rule (the `usedforsecurity=False` parameter and the `lgtm`+ # comment marker both failed to suppress it). The cache is+ # in-memory, so changing the hash function on upgrade has no+ # persistence-side effect — the same content always hashes+ # deterministically under whichever function is in use.+ hash_key = hashlib.sha256(original.encode()).hexdigest()[:24]entry = CompressionEntry(hash=hash_key,diff --git a/plugins/headroom-agent-hooks/.claude-plugin/plugin.json b/plugins/headroom-agent-hooks/.claude-plugin/plugin.jsonindex 4d8e4f107..d5f65b6ab 100644--- a/plugins/headroom-agent-hooks/.claude-plugin/plugin.json+++ b/plugins/headroom-agent-hooks/.claude-plugin/plugin.json@@ -1,6 +1,6 @@{"name": "headroom",- "version": "0.20.16",+ "version": "0.20.28","description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.","author": {"name": "Headroom Contributors",diff --git a/plugins/headroom-agent-hooks/.github/plugin/plugin.json b/plugins/headroom-agent-hooks/.github/plugin/plugin.jsonindex f91358182..2a5588a0b 100644--- a/plugins/headroom-agent-hooks/.github/plugin/plugin.json+++ b/plugins/headroom-agent-hooks/.github/plugin/plugin.json@@ -1,6 +1,6 @@{"name": "headroom",- "version": "0.20.16",+ "version": "0.20.28","description": "Headroom startup hooks for Claude Code and GitHub Copilot CLI.","author": {"name": "Headroom Contributors",