membrick 0.6.0__tar.gz

membrick-0.6.0/.claude/skills/membrick-memory/SKILL.md

@@ -0,0 +1,64 @@
+---
+name: membrick-memory
+description: Use the Membrick MCP server as the first source for project memory, prior decisions, and earlier work before inspecting the repository.
+---
+
+# Membrick Memory
+
+## MCP-first memory policy
+
+For requests about project history, prior decisions, architectural choices,
+earlier work, or "why" something was implemented, the first tool call must be
+`mcp__membrick__search`. Do not begin with `Bash`, `Read`, repository search, or
+another memory system for those requests.
+
+Use this sequence:
+
+1. Call `mcp__membrick__search` with a concise query describing the missing
+   context.
+2. Use the returned results as context and distinguish retrieved facts from
+   current source facts.
+3. Check cited files only after the MCP search, and only when the answer also
+   requires current source details.
+4. Prefer exact-revision or clearly relevant results and state uncertainty when
+   the results are incomplete.
+
+Use `mcp__membrick__info` only to check MCP workspace status or diagnose the
+server. It is not a memory search and should not replace `search`.
+
+## Explicit session recording
+
+Session recording is off by default until the user enables Membrick. Enabling
+Membrick also starts recording by default. Users can explicitly override this
+with `mcp__membrick__record_session` and `{"action":"stop"}`. Recording is
+capture-only: it records redacted lifecycle events locally and does not inject
+memory into the session.
+
+When the user asks to stop recording, call the same tool with
+`{"action":"stop"}`. Use `{"action":"status"}` only when the user asks
+whether recording is active. Never start recording implicitly for ordinary
+memory searches or project work.
+
+## Membrick memory controls
+
+When the user explicitly asks to turn Membrick memory off, call
+`mcp__membrick__disable_membrick`. When the user asks to turn it back on, call
+`mcp__membrick__enable_membrick`. These controls affect future Membrick retrieval
+and memory capture. Enabling memory also starts recording by default, while
+disabling memory leaves the recording choice unchanged. Use
+`mcp__membrick__membrick_status` when the user asks whether memory behavior is
+enabled. Do not toggle memory implicitly during a task.
+
+When installed, Claude Code may show `Membrick:ON | Record:OFF` in its native
+persistent status line. This is informational only; use the MCP control tools
+to change either state. Preserve any user-defined Claude status line.
+
+Replay is a separate CLI operation, not an automatic MCP action. Use
+`membrick --replay-session <session-id> --session-provider claude
+--replay-enable-membrick` when the user explicitly asks to rerun a recorded
+prompt with Claude. Replay always records its output and creates a new session
+ID; it does not modify the baseline archive.
+
+For requests that are exclusively about the current contents of a file or
+current command output, repository tools may be used directly. Do not claim
+that MCP was searched when the call was unavailable or returned no results.

membrick-0.6.0/.dockerignore

@@ -0,0 +1,17 @@
+.git
+.github
+.agents
+.claude
+.codex
+.history
+.membrick
+.membrick-cache
+target
+benchmark
+docs
+tests
+*.md
+!README.md
+.env
+.env.*
+!.env.example

membrick-0.6.0/.env.example

@@ -0,0 +1,5 @@
+# Server access token. Replace this with a long, random secret.
+SERVER_TOKEN=change-me-to-a-long-random-token
+
+# The server listens on 127.0.0.1:8080 by default when run locally.
+# Docker deployments should bind the container to 0.0.0.0:8080.

membrick-0.6.0/.gitignore

@@ -0,0 +1,94 @@
+# Generated by Cargo
+# will have compiled files and executables
+debug
+target
+
+# These are backup files generated by rustfmt
+**/*.rs.bk
+
+# MSVC Windows builds of rustc generate these, which store debugging information
+*.pdb
+
+# Generated by cargo mutants
+# Contains mutation testing data
+**/mutants.out*/
+
+# RustRover
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+# Codex and release artifacts
+.codex
+target/release
+
+# Membrick local cache
+.membrick-cache/
+.membrick/
+.lint-ai-cache/
+.lint-ai/
+
+# Local workspace file
+membrick.code-workspace
+lint-ai.code-workspace
+
+# Generated graph artifacts
+membrick-graph.dot
+membrick-graph.png
+membrick-graph.svg
+membrick-graph.html
+membrick-chunk-graph.html
+lint-ai-graph.dot
+lint-ai-graph.png
+lint-ai-graph.svg
+lint-ai-graph.html
+lint-ai-chunk-graph.html
+
+# Optional for library-style workflows
+!bindings/node/Cargo.lock
+
+# Local Python environments and caches
+.venv/
+
+# Local secrets
+.env
+/secrets/
+__pycache__/
+*.py[cod]
+
+# Local macOS/log artifacts
+.DS_Store
+build.log
+site/
+
+# Generated benchmark smoke outputs
+benchmark/*_results.json
+benchmark/codex_code/results/*
+!benchmark/codex_code/results/.gitkeep
+benchmark/agy/results/*
+!benchmark/agy/results/.gitkeep
+# Benchmark corpora and generated results are local-only. Ignore the directory
+# rather than individual filenames: untracked files here are picked up by
+# `cargo package` and would otherwise ship inside the published crate.
+benchmark/data/
+benchmark/claude_code/results/*
+!benchmark/claude_code/results/.gitkeep
+
+# Local data and scratch output from memory and benchmark runs
+data/memory-index/
+.benchmark-tmp/
+.codex-build/
+.codex-output/
+.membrick-agy-live.*/
+metrics-sess-*/
+# Keep raw copies of published LongMemEval reports out of Git; commit the .json.gz artifacts.
+comparison/results/retrieval-longmemeval-current-*.json
+comparison/results/segment-longmemeval-500-2026-10-*.json
+comparison/results/retrieval-longmemeval-route-width-*-100-2026-10-04.json
+comparison/results/segment-multisession-2026-10-04.json
+# Local provider benchmark runs contain large worktrees, logs, and binaries.
+/scripts/provider-performance-0.4.0/
+/comparison/results/release-v0.5.0-multisession-133-nobekind-c8fef3d7.json
+/comparison/results/release-v0.5.0-temporal-133-nobekind-c8fef3d7.json

membrick-0.6.0/AGENTS.md

@@ -0,0 +1,25 @@
+<!-- membrick:memory-policy:start -->
+## Membrick memory policy
+
+For requests about project history, prior decisions, architectural choices,
+earlier work, or "why" something was implemented, call the `membrick` MCP
+server's `search` tool before reading files or searching the repository.
+
+1. Call `search` with a concise query describing the missing context.
+2. Read the returned sources before opening files yourself.
+3. Say plainly when memory returned nothing, and only then fall back to
+   reading the repository.
+
+Recorded sessions are memory of this project, not documentation of it: a
+finding can be out of date. Check a conclusion against the code before relying
+on it, and prefer the most recent one when two disagree.
+<!-- membrick:memory-policy:end -->
+
+## User preferences
+
+Prefer tool-native configuration over environment variables when a clean,
+supported alternative exists. For Python extension packaging, use Maturin's
+supported workflow and let it manage PyO3 extension linker settings.
+
+In written responses, avoid `---` separators and keep punctuation and phrasing
+simple.

membrick-0.6.0/CITATION.cff

@@ -0,0 +1,11 @@
+cff-version: 1.2.0
+message: "If you use Membrick in research or software, please cite it as below."
+title: "Membrick: Persistent Memory for AI Coding Agents"
+type: software
+authors:
+  - name: RooAGI
+version: 0.6.0
+date-released: 2026-10-13
+repository-code: "https://github.com/RooAGI/Membrick"
+url: "https://www.rooagi.com"
+license: Apache-2.0