# inspeximus - Doramagic AI Context Pack

> Positioning: a pre-install experience and judgment asset. It helps the host AI get off to a good start, but it does not mean the project has already been installed, run, or validated.

## Sufficiency Principle

- **Sufficiency over compression**: The AI Context Pack should be sufficient for the host AI to understand the project's value, capability boundaries, entrypoints, risks, and evidence sources before starting work; it may be layered, but it does not aim for the shortest possible summary.
- **Compression policy**: Compress only noise and duplication, never context that affects judgment or the quality of the work.

## How the Host AI Should Use This

You are reading the AI Context Pack that Doramagic compiled for inspeximus. Treat it as pre-work context: help the user understand who it fits, what it can do, how to start, what must be verified after install, and where the risks are. Do not claim that you have already installed, run, or executed the target project.

## Claim Consumption Rules

- **Fact source**: Repo Evidence + Claim/Evidence Graph; the Human Wiki only supplies salience, terminology, and narrative structure.
- **Minimum status for a fact**: `supported`
- `supported`: May be used as a project fact, but the answer must cite the claim_id and evidence path.
- `weak`: Usable only as a low-confidence lead; the user must be asked to keep verifying.
- `inferred`: Usable only for risk notes or open questions; must not be packaged as a project fact.
- `unverified`: Must not be used as fact; state clearly that evidence is insufficient.
- `contradicted`: Must show the conflicting sources and must not force a single version on the user's behalf.

## Who It Fits Best

- **Developers already using host AIs such as Claude/Codex/Cursor/Gemini**: The README or plugin config mentions multiple host AIs. Evidence: `README.md` Claim: `clm_0003` supported 0.86

## What It Can Do

- **Multi-Host Install and Distribution** (Verify after install): The project contains plugin or marketplace configuration, indicating it targets install and distribution across one or more AI hosts. Evidence: `.claude-plugin/marketplace.json`, `.claude-plugin/plugin.json` Claim: `clm_0001` supported 0.86
- **Command-Line Startup or Install Flow** (Verify after install): The project documentation contains runnable commands; real use requires running them in a local or host environment. Evidence: `README.md`, `packages/adk-inspeximus/README.md`, `packages/langgraph-checkpoint-inspeximus/README.md`, `packages/langgraph-store-inspeximus/README.md` Claim: `clm_0002` supported 0.86

## How to Start

- `/plugin marketplace add DanceNitra/inspeximus` Evidence: `README.md` Claim: `clm_0004` supported 0.86
- `/plugin install inspeximus@inspeximus` Evidence: `README.md` Claim: `clm_0005` supported 0.86
- `pip install inspeximus          # zero required dependencies` Evidence: `README.md` Claim: `clm_0006` supported 0.86
- `claude mcp add inspeximus -e INSPEXIMUS_PATH=~/.inspeximus_memory.json -- uvx --from "inspeximus[mcp]" inspeximus-mcp` Evidence: `README.md` Claim: `clm_0007` supported 0.86
- `pip install "inspeximus[mcp]"     # the library + the MCP server SDK` Evidence: `README.md` Claim: `clm_0008` supported 0.86
- `pip install adk-inspeximus` Evidence: `packages/adk-inspeximus/README.md` Claim: `clm_0009` supported 0.86
- `pip install langgraph-checkpoint-inspeximus` Evidence: `packages/langgraph-checkpoint-inspeximus/README.md` Claim: `clm_0010` supported 0.86
- `pip install langgraph-store-inspeximus` Evidence: `packages/langgraph-store-inspeximus/README.md` Claim: `clm_0011` supported 0.86

## Continue-or-Stop Decision Card

- **Current recommendation**: Trial role matching first
- **Why**: This project is more of a role library; the core risk is picking the wrong role or treating role copy as execution capability. Trial role matching with Prompt Preview first, then decide whether to sandbox-import it.

### 30-Second Read

- **What to do now**: Trial role matching first
- **Minimum safe next step**: Trial role matching with Prompt Preview first; import in isolation only once satisfied
- **Do not trust yet**: Role quality and task fit cannot be trusted directly.
- **Continuing will touch**: Role selection bias, Command execution, Host AI configuration

### What You Can Trust Now

- **Target-audience signal: Developers already using host AIs such as Claude/Codex/Cursor/Gemini** (supported): Backed by a supported claim or project evidence, but that still is not the same as real install results. Evidence: `README.md` Claim: `clm_0003` supported 0.86
- **Capability exists: Multi-Host Install and Distribution** (supported): You can trust that the project contains signals of this capability; whether it fits your specific task still needs trial or after-install verification. Evidence: `.claude-plugin/marketplace.json`, `.claude-plugin/plugin.json` Claim: `clm_0001` supported 0.86
- **Capability exists: Command-Line Startup or Install Flow** (supported): You can trust that the project contains signals of this capability; whether it fits your specific task still needs trial or after-install verification. Evidence: `README.md`, `packages/adk-inspeximus/README.md`, `packages/langgraph-checkpoint-inspeximus/README.md`, `packages/langgraph-store-inspeximus/README.md` Claim: `clm_0002` supported 0.86
- **There are Quick Start / install-command signals** (supported): You can trust that the docs mention a startup or install entrypoint; do not run it directly in your primary environment because of that. Evidence: `README.md` Claim: `clm_0004` supported 0.86

### What You Cannot Trust Yet

- **Role quality and task fit cannot be trusted directly.** (unverified): A role library proves there are many roles; it does not prove each one fits your specific task or that a role produces high-quality results.
- **Do not treat role copy as real execution capability.** (unverified): Before install you can only judge whether the role description and task profile match; you cannot prove it can complete the task inside the host AI.
- **Real output quality cannot be trusted before install.** (unverified): Prompt Preview can only show how it guides you; it cannot prove result quality in the real project.
- **Host AI version compatibility cannot be trusted before install.** (unverified): Host loading rules and version differences across Claude, Cursor, Codex, Gemini, and others must be verified in a real environment.
- **That it will not pollute your existing host AI's behavior cannot be trusted directly.** (inferred): Skill, plugin, and AGENTS/CLAUDE/GEMINI instructions may change the host AI's default behavior. Evidence: `.claude-plugin/marketplace.json`, `.claude-plugin/plugin.json`
- **Safe rollback cannot be assumed by default.** (unverified): Unless the project clearly provides uninstall and recovery instructions, verify in an isolated environment first.
- **After a real install, is it compatible with the user's current host AI version?** (unverified): Compatibility can only be verified in the actual host environment. Evidence: `.claude-plugin/marketplace.json`, `.claude-plugin/plugin.json`
- **Does the project's output quality meet the user's specific task?** (unverified): The pre-install preview can only show flow and boundaries; it cannot replace real evaluation.

### What Continuing Will Touch

- **Role selection bias**: The user's judgment about which expert role should handle the task. Why: Picking the wrong role makes the AI answer from the wrong expert perspective, wasting time or misleading decisions.
- **Command execution**: Package managers, network downloads, the local plugin directory, project config, or the user's home directory. Why: Running the very first command can already change your environment; decide whether it is worth running first. Evidence: `README.md`, `packages/adk-inspeximus/README.md`, `packages/langgraph-checkpoint-inspeximus/README.md`, `packages/langgraph-store-inspeximus/README.md`
- **Host AI configuration**: The plugin, Skill, or rule-loading config of hosts like Claude/Codex/Cursor/Gemini/OpenCode. Why: Host configuration changes how the AI works afterward and may conflict with the user's existing rules. Evidence: `.claude-plugin/marketplace.json`, `.claude-plugin/plugin.json`
- **Local environment or project files**: Install results, plugin caches, project config, or local dependency directories. Why: The write scope and rollback path cannot be proven before install and need isolated verification. Evidence: `.claude-plugin/marketplace.json`, `.claude-plugin/plugin.json`, `README.md`, `packages/adk-inspeximus/README.md` et al.
- **Host AI context**: The AI Context Pack, Prompt Preview, Skill routing, risk rules, and project facts. Why: Importing context affects the host AI's later judgment, so avoid packaging unverified items as facts.

### Minimum Safe Next Steps

- **Run Prompt Preview first**: Use an interactive trial to verify the task profile and role match first; do not import the whole role library up front. (applies when: Applies to any project, especially when output quality is unknown.)
- **Trial-install only in an isolated directory or a test account**: Avoid letting install commands pollute your primary host AI, real projects, or home directory. (applies when: When there are signals of command execution, plugin config, or local writes.)
- **Back up your host AI configuration first**: Skill, plugin, and rule files may change the default behavior of Claude/Cursor/Codex. (applies when: When there is a plugin manifest, a Skill, or a host rule entrypoint.)
- **After install, verify just one minimal task**: Verify loading, compatibility, output quality, and rollback first, then decide whether to use it deeply. (applies when: When moving from a trial into a real workflow.)

### Exit Plan

- **Preserve the pre-install state**: Record the original host config and project state so you can later judge whether it is recoverable.
- **Be ready to remove the host plugin / Skill / rule entrypoint**: If behavior is off after the trial install, you can restore the host AI to its pre-trial state.
- **Keep a record of the original role selection**: If output goes off-topic, you can return to the task-profiling stage and reselect a role instead of pushing on with the wrong one.
- **Record the install commands and written paths**: Without clear uninstall instructions, you at least need to know which directories or configs to clean up manually.
- **If there is no rollback path, do not enter your primary environment**: No rollback is a blocker before continuing; do not proceed on trust or luck.

## What Can Only Be Previewed

- Explain who the project fits and what it can do
- Demonstrate a typical conversation flow based on project docs
- Help the user decide whether it is worth installing or researching further

## What Must Be Verified After Install

- Actually installing the Skill, plugin, or CLI
- Running scripts, modifying local files, or accessing external services
- Verifying real output quality, performance, and compatibility

## Boundary & Risk Decision Card

- **Mistaking the pre-install preview for a real run**: The user may overestimate how much configuration, permission, and compatibility verification the project has already done. Mitigation: Clearly separate prompt_preview_can_do from runtime_required. Claim: `clm_0012` inferred 0.45
- **Host AI plugin or Skill rule conflicts**: New rules may change how the user's existing host AI behaves. Mitigation: Inspect the plugin manifest and Skill files before installing, and test in isolation if needed. Evidence: `.claude-plugin/marketplace.json`, `.claude-plugin/plugin.json` Claim: `clm_0013` supported 0.86
- **Command execution will modify the local environment**: Install commands may write to the user's home directory, the host plugin directory, or project configuration. Mitigation: Run in an isolated environment or a test account first. Evidence: `README.md`, `packages/adk-inspeximus/README.md`, `packages/langgraph-checkpoint-inspeximus/README.md`, `packages/langgraph-store-inspeximus/README.md` Claim: `clm_0014` supported 0.86
- **To confirm**: After a real install, is it compatible with the user's current host AI version?. Why: Compatibility can only be verified in the actual host environment.
- **To confirm**: Does the project's output quality meet the user's specific task?. Why: The pre-install preview can only show flow and boundaries; it cannot replace real evaluation.
- **To confirm**: Do the install commands require network access, permissions, or global writes?. Why: This affects install risk in both enterprise and personal environments.

## Pre-Work Working Context

### Loading Order

- First read how_to_use.host_ai_instruction to establish the boundaries of this pre-install judgment asset.
- Read claim_graph_summary to confirm facts come from the Claim/Evidence Graph, not the Human Wiki narrative.
- Then read intended_users, capabilities, and quick_start_candidates to judge whether the user is a match.
- When you need to carry out a concrete task, check role_skill_index first, then evidence_index.
- For real install, file modification, network access, performance, or compatibility questions, turn to risk_card and boundaries.runtime_required.

### Task Routes

- **Multi-Host Install and Distribution**: State that this is an after-install capability first, then give a pre-install checklist. Boundary: Must be verified after a real install or run. Evidence: `.claude-plugin/marketplace.json`, `.claude-plugin/plugin.json` Claim: `clm_0001` supported 0.86
- **Command-Line Startup or Install Flow**: State that this is an after-install capability first, then give a pre-install checklist. Boundary: Must be verified after a real install or run. Evidence: `README.md`, `packages/adk-inspeximus/README.md`, `packages/langgraph-checkpoint-inspeximus/README.md`, `packages/langgraph-store-inspeximus/README.md` Claim: `clm_0002` supported 0.86

### Context Scale

- Total files: 255
- Important-file coverage: 40/255
- Evidence index entries: 79
- Role / Skill entries: 22

### Handling Insufficient Evidence

- **missing_evidence**: State that evidence is insufficient and ask the user for the target file, a README section, or after-install verification records; do not fill in facts.
- **out_of_scope_request**: State that the task is beyond the current AI Context Pack's evidence scope and suggest the user check the Human Manual or verify after a real install.
- **runtime_request**: Provide a pre-install checklist and command sources, but do not run commands for the user or claim they have been run.
- **source_conflict**: Show the conflicting sources side by side, mark them as unverified, and do not force a single version.

## Prompt Recipes

### Fit assessment

- Goal: Judge whether this project fits the user's current task.
- Expected output: A fit conclusion, key reasons, evidence citations, what can be previewed before install, what must be verified after install, and a next-step recommendation.

```text
Based on the AI Context Pack for inspeximus, ask me 3 necessary questions first, then judge whether it fits my task. The answer must cover: who it fits, what it can do, what it cannot do, whether it is worth installing, and where the evidence comes from. Every project fact must cite evidence_refs, source_paths, or a claim_id.
```

### Pre-install experience

- Goal: Let the user feel the core workflow before installing, while avoiding packaging the preview as real capability or a marketing promise.
- Expected output: An experience script with boundary labels, an after-install verification checklist, and a cautious recommendation; with no real-run promises or strong marketing language.

```text
Treat inspeximus as a pre-install experience asset, not an already-installed tool or a real runtime environment.

Output exactly four parts:
1. Ask me 3 necessary questions first.
2. Give an "experience script": use the three labels [Previewable before install], [Must verify after install], and [Insufficient evidence] to show how it might guide the workflow.
3. Give an after-install verification checklist: list which capabilities can only be confirmed after a real install, real host loading, and a real project run.
4. Give a cautious recommendation: only "worth researching/trialing further", "add information before deciding", or "not recommended to continue"; do not endorse the project.

Hard boundaries:
- Do not claim you have installed, run, executed tests, modified files, or produced real results.
- Do not write promise-like phrasing such as "auto-adapts", "guarantees passing", "perfect fit", or "strongly recommend installing".
- If you describe how it works after install, you must use a conditional such as "if installed successfully and the host loads the Skill correctly, it might...".
- The experience script may only be written as "example lines / hypothetical flow": use "might ask / might suggest / might show", not "has written, has generated, has passed, is running, is generating".
- Prompt Preview does not hand out install commands; if the user is ready to trial, only prompt them to read Quick Start and the Risk Card first and to verify in an isolated environment.
- Every project fact must come from a supported claim, evidence_refs, or source_paths; inferred/unverified items can only be risks or open questions.

```

### Role / Skill selection

- Goal: Pick the best-matching asset from the project's roles or Skills.
- Expected output: A list of candidate roles or Skills, each with an applicable scenario, evidence paths, risk boundary, and whether after-install verification is needed.

```text
Read role_skill_index and recommend 3-5 of the most relevant roles or Skills for my target task. For each recommendation, state the applicable scenario, likely output, risk boundary, and evidence_refs.
```

### Risk pre-check

- Goal: Identify environment, permission, rule-conflict, and quality risks before installing or adopting.
- Expected output: A checklist of environment, permission, dependency, license, host-conflict, quality risk, and unknown items.

```text
Based on risk_card, boundaries, and quick_start_candidates, give me a pre-install risk pre-check list. Do not run commands for me; only explain what I should check, why, and what impact a failure would have.
```

### Host AI kickoff instruction

- Goal: Turn the project context into a host AI instruction for the start of a conversation.
- Expected output: A pre-work instruction with clear boundaries and clear evidence citations, suitable to copy to a host AI.

```text
Based on the AI Context Pack for inspeximus, generate a pre-work instruction I can paste to my host AI. This instruction must obey not_runtime=true and must not claim the project has been installed, run, or produced real results.
```

## Role / Skill Index

- Indexed 22 role / Skill / project-doc entries.

- **inspeximus** (project_doc): "We have inspected" — the medieval charter that recites an earlier one word for word and attests it unaltered. The self-correcting memory layer for AI agents. Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `README.md`
- **MemoryAgentBench — Conflict Resolution, inspeximus eval** (project_doc): MemoryAgentBench — Conflict Resolution, inspeximus eval Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `bench/README.md`
- **inspeximus examples** (project_doc): Runnable, copy-paste examples. Each is self-contained and needs only pip install inspeximus . Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `examples/README.md`
- **inspeximus probes — runnable memory-reliability tests** (project_doc): inspeximus probes — runnable memory-reliability tests Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `probes/README.md`
- **inspeximus — launch site** (project_doc): A single-page marketing site for inspeximus https://github.com/DanceNitra/inspeximus : a zero-dependency memory layer and MCP server for AI agents. Dark, precise, with a scroll-driven 3D "memory ledger" narrative. Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `site/README.md`
- **MemOps: what happened when inspeximus's correction layer was measured against a keep-everything store** (project_doc): MemOps: what happened when inspeximus's correction layer was measured against a keep-everything store Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `benchmarks/memops/README.md`
- **n8n template — AI agent with self-correcting memory inspeximus** (project_doc): n8n template — AI agent with self-correcting memory inspeximus Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `examples/n8n/README.md`
- **adk-inspeximus** (project_doc): A Google ADK https://github.com/google/adk-python memory service backed by inspeximus https://github.com/DanceNitra/inspeximus . Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `packages/adk-inspeximus/README.md`
- **langgraph-checkpoint-inspeximus** (project_doc): A LangGraph checkpointer backed by inspeximus https://pypi.org/project/inspeximus/ . Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `packages/langgraph-checkpoint-inspeximus/README.md`
- **langgraph-store-inspeximus** (project_doc): A LangGraph BaseStore backed by inspeximus https://pypi.org/project/inspeximus/ — a drop-in for InMemoryStore that also remembers what a value used to be. Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `packages/langgraph-store-inspeximus/README.md`
- **The EU AI Act compliance-evidence layer for AI-agent memory** (project_doc): The EU AI Act compliance-evidence layer for AI-agent memory Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `docs/AI_ACT.md`
- **Security model** (project_doc): Honest scope: inspeximus has had a first internal security pass 2026-07-13 , not a formal third-party pentest. This document states what inspeximus defends, what it does not, and the known residual footguns — so you can decide what to rely on. If you find an issue, open a GitHub issue or security advisory. Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `SECURITY.md`
- **Use it from the shell: the inspeximus CLI 1.12.4** (project_doc): Use it from the shell: the inspeximus CLI 1.12.4 Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `docs/API.md`
- **inspeximus — Compliance Control Mapping** (project_doc): inspeximus — Compliance Control Mapping Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `docs/COMPLIANCE.md`
- **Framework integrations** (project_doc): inspeximus drops into the major agent frameworks as their native memory type , so an agent gets value-ranked recall plus correction-integrity a corrected fact does not resurrect on a later read without changing its code. Each adapter lives under inspeximus.integrations. and is an opt-in extra — import inspeximus stays zero-dependency, and the framework is imported lazily only when you use its adapter. Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `docs/INTEGRATIONS.md`
- **The second brain thinking layer** (project_doc): Optional add-on — a separate MCP server that reasons over a folder of Markdown notes. Click to expand. Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `docs/SECOND_BRAIN.md`
- **Changelog** (project_doc): All notable changes to inspeximus inspeximus . Format loosely follows Keep a Changelog; versioning is semver MAJOR = stable/breaking, MINOR = features, PATCH = fixes . Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `CHANGELOG.md`
- **inspeximus MCP — registry listing pack** (project_doc): inspeximus MCP — registry listing pack Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `MCP_LISTINGS.md`
- **Inspeximus** (project_doc): A persistent DocumentStore with a correction channel — supersede, revert, and receipted erasure Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `haystack_integration_inspeximus.md`
- **Agent-memory integrity benchmark open, cross-system, run-it-yourself** (project_doc): Agent-memory integrity benchmark open, cross-system, run-it-yourself Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `probes/INTEGRITY_BENCHMARK.md`
- **Probe spec — erasure completeness & revert exactness on MemOps data** (project_doc): Probe spec — erasure completeness & revert exactness on MemOps data Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `benchmarks/memops/ERASURE_REVERT_SPEC.md`
- **Pre-registration — inspeximus on MemOps internal diagnostic** (project_doc): Pre-registration — inspeximus on MemOps internal diagnostic Activation hint: Reference this when the user needs to understand the project's structure, install path, or boundaries. Evidence: `benchmarks/memops/PREREGISTRATION.md`

## Evidence Index

- Indexed 79 evidence entries.

- **inspeximus** (documentation): "We have inspected" — the medieval charter that recites an earlier one word for word and attests it unaltered. The self-correcting memory layer for AI agents. Evidence: `README.md`
- **MemoryAgentBench — Conflict Resolution, inspeximus eval** (documentation): MemoryAgentBench — Conflict Resolution, inspeximus eval Evidence: `bench/README.md`
- **inspeximus examples** (documentation): Runnable, copy-paste examples. Each is self-contained and needs only pip install inspeximus . Evidence: `examples/README.md`
- **inspeximus probes — runnable memory-reliability tests** (documentation): inspeximus probes — runnable memory-reliability tests Evidence: `probes/README.md`
- **inspeximus — launch site** (documentation): A single-page marketing site for inspeximus https://github.com/DanceNitra/inspeximus : a zero-dependency memory layer and MCP server for AI agents. Dark, precise, with a scroll-driven 3D "memory ledger" narrative. Evidence: `site/README.md`
- **MemOps: what happened when inspeximus's correction layer was measured against a keep-everything store** (documentation): MemOps: what happened when inspeximus's correction layer was measured against a keep-everything store Evidence: `benchmarks/memops/README.md`
- **n8n template — AI agent with self-correcting memory inspeximus** (documentation): n8n template — AI agent with self-correcting memory inspeximus Evidence: `examples/n8n/README.md`
- **adk-inspeximus** (documentation): A Google ADK https://github.com/google/adk-python memory service backed by inspeximus https://github.com/DanceNitra/inspeximus . Evidence: `packages/adk-inspeximus/README.md`
- **langgraph-checkpoint-inspeximus** (documentation): A LangGraph checkpointer backed by inspeximus https://pypi.org/project/inspeximus/ . Evidence: `packages/langgraph-checkpoint-inspeximus/README.md`
- **langgraph-store-inspeximus** (documentation): A LangGraph BaseStore backed by inspeximus https://pypi.org/project/inspeximus/ — a drop-in for InMemoryStore that also remembers what a value used to be. Evidence: `packages/langgraph-store-inspeximus/README.md`
- **The EU AI Act compliance-evidence layer for AI-agent memory** (documentation): The EU AI Act compliance-evidence layer for AI-agent memory Evidence: `docs/AI_ACT.md`
- **Package** (package_manifest): { "name": "inspeximus-site", "private": true, "version": "1.1.0", "type": "module", "scripts": { "dev": "vite", "build": "vite build", "preview": "vite preview" }, "dependencies": { "gsap": "^3.12.5", "lenis": "^1.1.14", "three": "^0.169.0" }, "devDependencies": { "typescript": "^5.6.3", "vite": "^5.4.10", "@types/three": "^0.169.0" } } Evidence: `site/package.json`
- **Marketplace** (structured_config): { "name": "inspeximus", "owner": { "name": "Rastislav Drahos", "url": "https://github.com/DanceNitra" }, "metadata": { "description": "Self-correcting memory for coding agents — corrections stick, deletions are provable, and nothing calls a model on the write path.", "version": "1.0.0" }, "plugins": { "name": "inspeximus", "source": "./", "description": "Persistent memory for Claude Code that corrects itself: write a new value for a fact and the old one is retired, revert puts it back, and forget leaves a signed, content-free receipt. Deterministic and zero-dependency — no LLM call when storing, so the same input always produces the same store.", "version": "1.25.0", "author": { "name": "Ra… Evidence: `.claude-plugin/marketplace.json`
- **Plugin** (structured_config): { "name": "inspeximus", "version": "1.25.0", "description": "Self-correcting memory for Claude Code. Facts you correct stay corrected, deletions leave a signed receipt, and storing something never calls a model.", "author": { "name": "Rastislav Drahos", "url": "https://github.com/DanceNitra" }, "homepage": "https://github.com/DanceNitra/inspeximus", "repository": "https://github.com/DanceNitra/inspeximus", "license": "MIT", "keywords": "memory", "mcp", "agent-memory", "supersession", "erasure", "provenance" , "mcpServers": { "inspeximus": { "command": "uvx", "args": "--from", "inspeximus mcp ", "inspeximus-mcp" , "env": { "INSPEXIMUS PATH": "${CLAUDE PROJECT DIR}/.inspeximus/memory.json" }… Evidence: `.claude-plugin/plugin.json`
- **License** (source_file): Copyright c 2026 Agora autonomous research organization Evidence: `LICENSE`
- **Security model** (documentation): Honest scope: inspeximus has had a first internal security pass 2026-07-13 , not a formal third-party pentest. This document states what inspeximus defends, what it does not, and the known residual footguns — so you can decide what to rely on. If you find an issue, open a GitHub issue or security advisory. Evidence: `SECURITY.md`
- **Use it from the shell: the inspeximus CLI 1.12.4** (documentation): Use it from the shell: the inspeximus CLI 1.12.4 Evidence: `docs/API.md`
- **inspeximus — Compliance Control Mapping** (documentation): inspeximus — Compliance Control Mapping Evidence: `docs/COMPLIANCE.md`
- **Framework integrations** (documentation): inspeximus drops into the major agent frameworks as their native memory type , so an agent gets value-ranked recall plus correction-integrity a corrected fact does not resurrect on a later read without changing its code. Each adapter lives under inspeximus.integrations. and is an opt-in extra — import inspeximus stays zero-dependency, and the framework is imported lazily only when you use its adapter. Evidence: `docs/INTEGRATIONS.md`
- **The second brain thinking layer** (documentation): Optional add-on — a separate MCP server that reasons over a folder of Markdown notes. Click to expand. Evidence: `docs/SECOND_BRAIN.md`
- **Claims Audit** (source_file): PKG ENV = "INSPEXIMUS AUDIT PKG" ⋮---- def load ⋮---- p = os.environ.get PKG ENV ⋮---- def store tmp name, keyed=True ⋮---- d = pathlib.Path tempfile.mkdtemp prefix=f"audit {tmp name} " m = Inspeximus path=str d / "store.jsonl" ⋮---- def c zero deps ⋮---- root = pathlib.Path inspeximus. file .resolve .parent meta = list root.parent.glob " .dist-info/METADATA" requires = ⋮---- core = root / "core.py" ⋮---- def c no llm on write ⋮---- real = socket.socket ⋮---- class Blocked socket.socket ⋮---- def init self, a, k ⋮---- hits = m.recall "address", k=2, mode="lexical", reinforce=False ⋮---- def c supersession ⋮---- """README: 'corrections stick — supersession + echo guard'.""" ⋮---- rows = r "s… Evidence: `claims_audit.py`
- **01 Basics** (source_file): m = Inspeximus "memory.json" Evidence: `examples/01_basics.py`
- **07 Witness Pool** (source_file): STORE ID = "acme-prod-memory" ⋮---- def main ⋮---- m = Inspeximus path=None, receipts=True ⋮---- anchor = m.anchor ⋮---- witnesses = Witness for in range 3 allow = w.public for w in witnesses ⋮---- out = collect cosignatures STORE ID, anchor, witnesses v = Inspeximus.verify cosigned anchor anchor, out "cosignatures" , allow, threshold=2 ⋮---- anchor2 = m.anchor out2 = collect cosignatures STORE ID, anchor2, witnesses ⋮---- forked = dict anchor2 ; forked "writes tip" = "f0rged" + anchor2 "writes tip" 6: ⋮---- out3 = collect cosignatures STORE ID, forked, witnesses ⋮---- v3 = Inspeximus.verify cosigned anchor forked, out3 "cosignatures" , allow, threshold=2 Evidence: `examples/07_witness_pool.py`
- **--- after the agent rewrites using the replacements, it comes back clean ---** (source_file): def main ⋮---- mem = Inspeximus path=None ⋮---- generated = hits = check code mem, generated ⋮---- --- after the agent rewrites using the replacements, it comes back clean --- fixed = Evidence: `examples/08_code_guard.py`
- **09 Audit Bundle** (source_file): def main ⋮---- m = Inspeximus path=None, receipts=True ⋮---- bundle = build bundle m ⋮---- res = verify bundle bundle ⋮---- forged = dict bundle ⋮---- res2 = verify bundle forged Evidence: `examples/09_audit_bundle.py`
- **1. gone from the store** (source_file): PKG ENV = "INSPEXIMUS AUDIT PKG" ⋮---- SCENARIOS = ⋮---- def load ⋮---- p = os.environ.get PKG ENV ⋮---- def secrets sc ⋮---- out = ⋮---- def state hash m ⋮---- def run scenario sc, run idx ⋮---- """One full build → erase → verify cycle. Returns checks: dict str, bool , evidence: list str .""" ⋮---- d = pathlib.Path tempfile.mkdtemp prefix=f"gov {run idx} " ⋮---- m = Inspeximus path=str d / "store.jsonl" , receipts=True ⋮---- owned ids = m.remember t, source={"doc": sc "subject" } for t in sc "owned" ⋮---- n before = len m.items ⋮---- res = {"erased": 0, "ids": , "tombstones": 0} ⋮---- res = m.forget subject sc "subject" , request id=f"gdpr-{run idx}", ⋮---- secrets = secrets sc ⋮---- 1. go… Evidence: `governance_audit.py`
- **Init** (source_file): all = Evidence: `inspeximus/__init__.py`
- **Audit Bundle** (source_file): BUNDLE KIND = "inspeximus.audit bundle/1" ⋮---- def bundle hash bundle: dict - str ⋮---- def content free writes store - list ⋮---- def content free tombstones store - list ⋮---- out = ⋮---- rec = {k: t.get k for k in "seq", "memory id", "ts", "request id", "prev", "hash" } ⋮---- def build bundle store, expected pubkey: str None = None, sign=None - dict ⋮---- anchor = store.anchor sign=sign bundle = { ⋮---- def rewalk records: list, kind: str - tuple str, int ⋮---- prev = GENESIS ⋮---- core = Inspeximus. chain core r, kind ⋮---- prev = r.get "hash" ⋮---- def verify bundle bundle: dict, witnesses: list None = None, threshold: int = 1 - dict ⋮---- def ok msg : checks.append msg def bad msg :… Evidence: `inspeximus/audit_bundle.py`
- **audit-verify needs only the bundle file — never open a store that would create one as a side effect .** (source_file): def embedder ⋮---- url = os.environ.get "INSPEXIMUS EMBED URL", "" .strip ⋮---- model = os.environ.get "INSPEXIMUS EMBED MODEL", "text-embedding-3-small" .strip key = os.environ.get "INSPEXIMUS EMBED KEY", "" .strip ⋮---- def embed text: str ⋮---- body = json.dumps {"model": model, "input": text} .encode headers = {"Content-Type": "application/json"} ⋮---- req = urllib.request.Request url, data=body, headers=headers ⋮---- def store path, persist vectors: bool = False, receipts: bool = False ⋮---- p = path or os.environ.get "INSPEXIMUS PATH" or "inspeximus memory.json" ⋮---- receipts = True ⋮---- st = Inspeximus path=p, embed= embedder , persist vectors=persist vectors, receipts=receipts ⋮--… Evidence: `inspeximus/cli.py`
- **Code Guard** (source_file): PREFIX = "code::symbol::" ⋮---- def key name: str - str ⋮---- def reason from rec: dict - str ⋮---- t = rec.get "text" or "" ⋮---- def deprecate symbol store, old: str, new: str, reason: str = "" - dict ⋮---- """Record that code symbol old was replaced by new a keyed supersession -- deterministic, no LLM . old / new are identifiers as they appear in code old fn , Client.connect , LEGACY FLAG . A later deprecation of the same old supersedes the replacement. Returns the recorded deprecation.""" old = str old .strip new = str new .strip ⋮---- text = f"{old} was replaced by {new}" + f": {reason}" if reason else "" ⋮---- def symbol status store, name: str - dict ⋮---- rec = store. current active… Evidence: `inspeximus/code_guard.py`
- **Compliance** (source_file): CONTROLS = ⋮---- def compliance report store, expected pubkey: str None = None - dict ⋮---- anchor = store.anchor gov = store.governance report expected pubkey sup = store.supersession report n writes = anchor.get "n writes" or 0 n tomb = anchor.get "n tombstones" or 0 n sup = sup.get "superseded total" or 0 receipts on = bool getattr store, "receipts enabled", False ⋮---- live = {"write receipts": n writes, "erasures": n tomb, "superseded": n sup} ⋮---- controls = ⋮---- count = live.get live key if live key else None ⋮---- status = "needs receipts" ⋮---- status = "available" ⋮---- status = "evidence" ⋮---- n receipts = len getattr store, " receipts", has content = any r.get "status" == "ac… Evidence: `inspeximus/compliance.py`
- **--- universal-executor detection 1.2.0 -------------------------------------------------------------------** (source_file): np = None ⋮---- HAVE ED = True ⋮---- HAVE ED = False ⋮---- HAVE AEAD = True ⋮---- HAVE AEAD = False ⋮---- INSPEXIMUS ENC MAGIC = b"INSP\x01" ⋮---- def new encryption key - bytes ⋮---- def derive key passphrase: str, salt: bytes - bytes ⋮---- def encrypt blob key: bytes, plaintext: bytes, salt: bytes - bytes ⋮---- nonce = os.urandom 12 header = INSPEXIMUS ENC MAGIC + salt + nonce ct = AESGCM key .encrypt nonce, plaintext, header ⋮---- def parse enc header blob: bytes ⋮---- def decrypt blob key: bytes, blob: bytes - bytes ⋮---- GENESIS = "0" 64 ⋮---- def canon obj - bytes ⋮---- def sha256 hex b: bytes - str ⋮---- def new receipt keypair ⋮---- sk = Ed25519SK.generate ⋮---- def new source keypa… Evidence: `inspeximus/core.py`
- **Deletion Manifest** (source_file): HAVE ED = True ⋮---- HAVE ED = False ⋮---- GENESIS = "0" 64 ⋮---- def canon obj - bytes ⋮---- def sha256 b - str ⋮---- class ErasureTarget ⋮---- name = "unnamed-target" ⋮---- def erase self, subject: str - dict ⋮---- def still recoverable self, subject: str, values - bool ⋮---- class DeletionManifest ⋮---- def init self, sign sk hex: str None = None, pubkey hex: str None = None ⋮---- def register self, target: ErasureTarget - "DeletionManifest" ⋮---- prev = GENESIS entries = ⋮---- res = t.erase subject or {} erased = int res.get "erased", 0 recoverable = bool t.still recoverable subject, values err = None ⋮---- e = {"target": t.name, "erased": erased, "still recoverable": recoverable, ⋮----… Evidence: `inspeximus/deletion_manifest.py`
- **Erasure Auditor** (source_file): class StoreProbe ⋮---- name = "unnamed-store" kind = "generic" ⋮---- def recover self, subject: str, values - dict ⋮---- class TextStoreProbe StoreProbe ⋮---- kind = "text" ⋮---- def init self, name: str, texts ⋮---- def recover self, subject, values ⋮---- blob = " \n ".join self. get or .lower hits = v for v in values if re.search r" ? 0 and frac 0 ⋮---- class ErasureAuditor ⋮---- def init self ⋮---- def register self, probe: StoreProbe - "ErasureAuditor" ⋮---- audit = self.audit subject, values, probe kwargs receipt = { ⋮---- def audit self, subject: str, values, probe kwargs - dict ⋮---- results = ⋮---- r = p.recover subject, values, {k: v for k, v in probe kwargs.items ⋮---- r = p.recov… Evidence: `inspeximus/erasure_auditor.py`
- **── GOVERNANCE / INTEGRITY tools the surface a serious buyer checks — previously absent from the MCP ──────** (source_file): def make embedders ⋮---- url = os.environ.get "INSPEXIMUS EMBED URL", "" .strip ⋮---- model = os.environ.get "INSPEXIMUS EMBED MODEL", "text-embedding-3-small" .strip key = os.environ.get "INSPEXIMUS EMBED KEY", "" .strip ⋮---- def embed text: str, prefix: str = "" ⋮---- body = json.dumps {"model": model, "input": prefix + text} .encode headers = {"Content-Type": "application/json"} ⋮---- req = urllib.request.Request url, data=body, headers=headers ⋮---- PATH = os.environ.get "INSPEXIMUS PATH", "inspeximus memory.json" ⋮---- RECEIPTS = os.environ.get "INSPEXIMUS RECEIPTS", "" .strip .lower in "1", "true", "yes", "on" ⋮---- MEM = Inspeximus PATH, embed= EMB DOC, embed query= EMB QUERY, embed… Evidence: `inspeximus/mcp_server.py`
- **Witness Pool** (source_file): def public from secret secret hex: str - str ⋮---- sk = Ed25519PrivateKey.from private bytes bytes.fromhex secret hex ⋮---- class Witness ⋮---- def init self, secret hex: str None = None, state path: str None = None ⋮---- public = public from secret secret hex ⋮---- def persist self - None ⋮---- d = os.path.dirname os.path.abspath self. state path or "." ⋮---- def cosign self, store id: str, anchor: dict - tuple str, str ⋮---- prior = self. last.get str store id sig = witness cosign self. secret, anchor, prior anchor=prior ⋮---- def last head self, store id: str - dict None ⋮---- def collect cosignatures store id: str, anchor: dict, witnesses - dict ⋮---- def http witness url: str, timeout:… Evidence: `inspeximus/witness_pool.py`
- **Witness Server** (source_file): def make handler witness: Witness ⋮---- class Handler BaseHTTPRequestHandler ⋮---- def send self, code: int, obj: dict ⋮---- body = json.dumps obj .encode ⋮---- def do GET self ⋮---- def do POST self ⋮---- n = int self.headers.get "Content-Length" or 0 body = json.loads self.rfile.read n or b"{}" ⋮---- def log message self, a : quiet by default ⋮---- w = Witness secret hex=secret hex, state path=state path httpd = ThreadingHTTPServer host, port , make handler w ⋮---- def main ⋮---- ap = argparse.ArgumentParser description="inspeximus reference witness server" ⋮---- a = ap.parse args Evidence: `inspeximus/witness_server.py`
- **Crewai** (source_file): class InspeximusStorage ComplianceMixin ⋮---- store = Inspeximus path=path, embed=embed, receipts=receipts ⋮---- def save self, value: Any, metadata: Optional Dict str, Any = None - None ⋮---- metadata = dict metadata or {} text = value if isinstance value, str else str value ⋮---- key = metadata.pop "key", None obj = metadata.pop "object", None tags = self. tag extra tags = metadata.pop "tags", None ⋮---- hits = self.store.recall query, k=limit or out: List Dict str, Any = ⋮---- score = h.get "score" ⋮---- def reset self - None ⋮---- ids = r "id" for r in list getattr self.store, "items", Evidence: `inspeximus/integrations/crewai.py`
- **Langchain** (source_file): class InspeximusRetriever BaseRetriever, ComplianceMixin ⋮---- k: int = 5 store: Any = None ⋮---- def get relevant documents self, query: str, , run manager=None - List Document ⋮---- hits = self.store.recall query, k=self.k or ⋮---- def add self, text: str, key: Optional str = None, kw: Any - None ⋮---- class InspeximusChatMessageHistory BaseChatMessageHistory, ComplianceMixin ⋮---- def init self, session id: str, path: str None = None, store: Any = None, embed=None ⋮---- @property def messages self - List BaseMessage ⋮---- rows = self.store.recall self. tag, k=1000, where={"tags": {"$contains": self. tag}} \ rows = sorted rows, key=lambda r: r.get "ts", 0 ⋮---- out = ⋮---- def add message… Evidence: `inspeximus/integrations/langchain.py`
- **Langgraph** (source_file): b = lambda x: base64.b64encode x .decode ub = lambda s: base64.b64decode s.encode ⋮---- def dt ts: float - datetime ⋮---- class InspeximusStore BaseStore, ComplianceMixin ⋮---- store = Inspeximus path=path, receipts=receipts ⋮---- @staticmethod def mkey namespace: tuple str, ... , key: str - str ⋮---- def active self, namespace, key ⋮---- mk = self. mkey namespace, key rows = r for r in self.store.items if r.get "status" == "active" and r.get "meta" or {} .get "mkey" == mk ⋮---- def to item self, rec - Item ⋮---- m = rec.get "meta" or {} ⋮---- def batch self, ops - list ⋮---- results: list = ⋮---- rec = self. active op.namespace, op.key ⋮---- mk = self. mkey op.namespace, op.key ⋮---- ids =… Evidence: `inspeximus/integrations/langgraph.py`
- **source= is what makes this message erasable by subject; see the langgraph adapter for** (source_file): class InspeximusMemoryBlock BaseMemoryBlock str , ComplianceMixin ⋮---- name: str = Field default="InspeximusMemory" description: Optional str = Field k: int = Field default=5, description="How many memories to inject." store: Any = PrivateAttr default=None subject: str = PrivateAttr default="llamaindex" ⋮---- store = Inspeximus path=path ⋮---- @property def store self ⋮---- @staticmethod def text msg: ChatMessage - str ⋮---- c = getattr msg, "content", None ⋮---- async def aput self, messages: List ChatMessage - None ⋮---- t = self. text m .strip ⋮---- source= is what makes this message erasable by subject; see the langgraph adapter for why the separator is "::" and not "/" a path-shaped s… Evidence: `inspeximus/integrations/llamaindex.py`
- **Init** (source_file): all = "InspeximusMemoryService", "register" Evidence: `packages/adk-inspeximus/adk_inspeximus/__init__.py`
- **Init** (source_file): all = "InspeximusSaver" version = "0.1.0" Evidence: `packages/langgraph-checkpoint-inspeximus/langgraph/checkpoint/inspeximus/__init__.py`
- **Init** (source_file): all = "InspeximusStore" version = "0.1.0" Evidence: `packages/langgraph-store-inspeximus/langgraph/store/inspeximus/__init__.py`
- **Changelog** (documentation): All notable changes to inspeximus inspeximus . Format loosely follows Keep a Changelog; versioning is semver MAJOR = stable/breaking, MINOR = features, PATCH = fixes . Evidence: `CHANGELOG.md`
- **inspeximus MCP — registry listing pack** (documentation): inspeximus MCP — registry listing pack Evidence: `MCP_LISTINGS.md`
- **Table of Contents** (documentation): Table of Contents - Overview overview - Installation installation - Usage usage - License license Evidence: `haystack_integration_inspeximus.md`
- **Agent-memory integrity benchmark open, cross-system, run-it-yourself** (documentation): Agent-memory integrity benchmark open, cross-system, run-it-yourself Evidence: `probes/INTEGRITY_BENCHMARK.md`
- **Probe spec — erasure completeness & revert exactness on MemOps data** (documentation): Probe spec — erasure completeness & revert exactness on MemOps data Evidence: `benchmarks/memops/ERASURE_REVERT_SPEC.md`
- **Pre-registration — inspeximus on MemOps internal diagnostic** (documentation): Pre-registration — inspeximus on MemOps internal diagnostic Evidence: `benchmarks/memops/PREREGISTRATION.md`
- **Control Rawiter** (structured_config): { "ok": true, "rows": { "row": 0, "facts": 455, "n": 50, "raw iterative": 0.14 }, { "row": 1, "facts": 2310, "n": 50, "raw iterative": 0.12 }, { "row": 2, "facts": 4580, "n": 50, "raw iterative": 0.14 }, { "row": 3, "facts": 18332, "n": 50, "raw iterative": 0.08 } } Evidence: `bench/_control_rawiter.json`
- **Diag Cr2 Result** (structured_config): { "ok": true, "rows": { "row": 0, "facts": 455, "keyed facts": 430, "keys": 276, "n": 30, "keyed top1": 0.0, "keyed top3": 0.133, "gold is a latest value": 0.967 }, { "row": 1, "facts": 2310, "keyed facts": 2144, "keys": 1360, "n": 30, "keyed top1": 0.033, "keyed top3": 0.067, "gold is a latest value": 0.9 }, { "row": 2, "facts": 4580, "keyed facts": 4243, "keys": 2673, "n": 30, "keyed top1": 0.033, "keyed top3": 0.1, "gold is a latest value": 0.967 } } Evidence: `bench/_diag_cr2_result.json`
- **Diag Cr Result** (structured_config): { "ok": true, "rows": { "row": 0, "facts": 455, "n": 30, "lex single": 0.433, "sem single": 0.433, "lex iter": 0.433, "sem iter": 0.533 }, { "row": 1, "facts": 2310, "n": 30, "lex single": 0.133, "sem single": 0.267, "lex iter": 0.167, "sem iter": 0.267 }, { "row": 2, "facts": 4580, "n": 30, "lex single": 0.267, "sem single": 0.3, "lex iter": 0.3, "sem iter": 0.367 } } Evidence: `bench/_diag_cr_result.json`
- **Results Cr Sweep** (structured_config): { "ok": true, "model": "deepseek-v4-flash", "hops": 2, "k": 15, "rows": { "row": 0, "facts": 455, "n": 50, "base full": 0.8, "inspeximus single": 0.28, "inspeximus iterative": 0.72 }, { "row": 1, "facts": 2310, "n": 50, "base full": 0.26, "inspeximus single": 0.08, "inspeximus iterative": 0.52 }, { "row": 2, "facts": 4580, "n": 50, "base full": 0.32, "inspeximus single": 0.06, "inspeximus iterative": 0.42 }, { "row": 3, "facts": 18332, "n": 50, "base full": "N/A context exceeds window ", "inspeximus single": 0.08, "inspeximus iterative": 0.38 } } Evidence: `bench/results_cr_sweep.json`
- **Property Benchmark.Result** (structured_config): { "n properties": 8, "passed": 8, "results": { "supersession returns current": { "metric": "current@1", "value": 1.0, "baseline": 0.0, "pass": true, "note": "inspeximus keyed vs naive keep-all" }, "revert restores prior": { "metric": "restored@1", "value": 1.0, "baseline": null, "pass": true, "note": "" }, "erasure reconstructibility": { "metric": "leak rate", "value": 0.0, "baseline": null, "pass": true, "note": "content gone from items + recall" }, "poison trusted only": { "metric": "truth@1", "value": 1.0, "baseline": 0.0, "pass": true, "note": "trusted only vs default recall, adaptive attacker" }, "noop dedup": { "metric": "new records on restate", "value": 0, "baseline": null, "pass":… Evidence: `benchmarks/property_benchmark.result.json`
- **Glama** (structured_config): { "$schema": "https://glama.ai/mcp/schemas/server.json", "maintainers": "DanceNitra" } Evidence: `glama.json`
- **Conflict Depth Compounding Result** (structured_config): { "ok": true, "model": "deepseek-v4-flash", "N": 30, "rows": { "hops": 1, "n": 30, "supersession": 1.0, "accumulate labeled": 1.0, "accumulate ordered": 1.0, "distractor": 1.0, "gap super minus ordered": 0.0 }, { "hops": 2, "n": 30, "supersession": 1.0, "accumulate labeled": 1.0, "accumulate ordered": 1.0, "distractor": 1.0, "gap super minus ordered": 0.0 }, { "hops": 3, "n": 30, "supersession": 1.0, "accumulate labeled": 0.967, "accumulate ordered": 1.0, "distractor": 1.0, "gap super minus ordered": 0.0 }, { "hops": 4, "n": 30, "supersession": 1.0, "accumulate labeled": 0.9, "accumulate ordered": 1.0, "distractor": 1.0, "gap super minus ordered": 0.0 } } Evidence: `probes/conflict_depth_compounding_result.json`
- **Cost Per Op Result** (structured_config): { "metric": "LLM-calls & tokens per op, by phase; extract 'site' shows where extraction is paid", "n": 10, "systems": { "inspeximus": { "extract": { "llm calls per op": 0.0, "tokens per op": 0.0, "site": "upstream/caller" }, "retain": { "llm calls per op": 0.0, "tokens per op": 0.0 }, "recall": { "llm calls per op": 0.0, "tokens per op": 0.0 } } } } Evidence: `probes/cost_per_op_result.json`
- **Generative Agents Agent Result** (structured_config): { "ok": true, "model": "deepseek-v4-flash", "scale": 1000, "k": 8, "rows": { "contra": 0.2, "scale": 1000, "n queries": 9, "errors": 0, "ga agent acc": 1.0, "inspeximus agent acc": 1.0 }, { "contra": 0.5, "scale": 1000, "n queries": 19, "errors": 0, "ga agent acc": 0.947, "inspeximus agent acc": 1.0 }, { "contra": 0.9, "scale": 1000, "n queries": 37, "errors": 0, "ga agent acc": 0.838, "inspeximus agent acc": 1.0 } } Evidence: `probes/generative_agents_agent_result.json`
- **Generative Agents Retrieval Stress Result** (structured_config): { "contra frac": 0.2, "scale": 0, "ga acc": 1.0, "inspeximus acc": 1.0, "ga stale ctx": 0.9, "inspeximus stale ctx": 0.0 }, { "contra frac": 0.2, "scale": 1000, "ga acc": 1.0, "inspeximus acc": 1.0, "ga stale ctx": 0.65, "inspeximus stale ctx": 0.0 }, { "contra frac": 0.2, "scale": 5000, "ga acc": 1.0, "inspeximus acc": 1.0, "ga stale ctx": 0.55, "inspeximus stale ctx": 0.0 }, { "contra frac": 0.5, "scale": 0, "ga acc": 0.9, "inspeximus acc": 1.0, "ga stale ctx": 0.728, "inspeximus stale ctx": 0.0 }, { "contra frac": 0.5, "scale": 1000, "ga acc": 0.845, "inspeximus acc": 1.0, "ga stale ctx": 0.834, "inspeximus stale ctx": 0.0 }, { "contra frac": 0.5, "scale": 5000, "ga acc": 0.845, "inspexi… Evidence: `probes/generative_agents_retrieval_stress_result.json`
- The remaining 19 evidence entries are in `AI_CONTEXT_PACK.json` or `EVIDENCE_INDEX.json`.

## Rules the Host AI Must Follow

- **Treat this asset as pre-work context, not a runtime environment.**: The AI Context Pack contains only an evidence-backed understanding of the project, not the project's executable state. Evidence: `README.md`, `bench/README.md`, `examples/README.md`
- **When answering the user, distinguish what can be previewed from what can only be verified after install.**: The consumer value of the pre-install experience comes from reducing bad installs and misjudgments, not from pretending to be a real run. Evidence: `README.md`, `bench/README.md`, `examples/README.md`

## Questions the User Should Answer First

- Which host AI or local environment do you plan to use it in?
- Do you just want to experience the workflow first, or are you ready to actually install?
- What matters most to you: install cost, output quality, or conflicts with your existing rules?

## Acceptance Checks

- Every capability claim can be traced back to a file path in evidence_refs.
- AI_CONTEXT_PACK.md does not package previews as a real run.
- The user can understand who it fits, what it can do, how to start, and the risk boundaries within 3 minutes.

---

## Doramagic Context Augmentation

The following sections strengthen the repository context for a host AI. Human Manual data is a reading route, and pitfall notes become operating constraints.

## Human Manual Outline

Usage rule: this is only a reading route and salience signal, not factual authority. Concrete claims must still return to repo evidence or Claim Graph.

Host AI hard rules:
- Do not treat page titles, section order, summaries, or importance values as factual project evidence.
- When explaining the Human Manual outline, state that it is only a reading route or salience signal.
- Capability, installation, compatibility, runtime state, and risk claims must cite repo evidence, source paths, or Claim Graph.

- **Introduction, Design Philosophy & Competitive Position**: importance `high`
  - source_paths: README.md, inspeximus/__init__.py, inspeximus/core.py, claims_audit.py, governance_audit.py
- **Core API & The Four Memory Operations**: importance `high`
  - source_paths: inspeximus/core.py, inspeximus/deletion_manifest.py, inspeximus/code_guard.py, inspeximus/audit_bundle.py, inspeximus/witness_pool.py
- **Governance, Verifiable Erasure & EU AI Act Compliance**: importance `high`
  - source_paths: inspeximus/erasure_auditor.py, inspeximus/deletion_manifest.py, inspeximus/audit_bundle.py, inspeximus/witness_pool.py, inspeximus/witness_server.py
- **Integrations, MCP Server, CLI & Framework Adapters**: importance `medium`
  - source_paths: inspeximus/mcp_server.py, inspeximus/cli.py, inspeximus/integrations/langgraph.py, inspeximus/integrations/langchain.py, inspeximus/integrations/crewai.py

## Repo Inspection Evidence

- repo_clone_verified: true
- repo_inspection_verified: true
- repo_commit: `24968aefec04591528a341f9600d4d9f54226eb4`
- inspected_files: `Dockerfile`, `README.md`, `pyproject.toml`, `docs/AI_ACT.md`, `docs/API.md`, `docs/COMPLIANCE.md`, `docs/INTEGRATIONS.md`, `docs/SECOND_BRAIN.md`, `examples/01_basics.py`, `examples/02_correction_and_erasure.py`, `examples/03_semantic_recall.py`, `examples/04_encryption.py`, `examples/05_review_trigger.py`, `examples/06_gdpr_erasure_receipt.py`, `examples/07_langgraph_memory.py`, `examples/07_witness_pool.py`, `examples/08_code_guard.py`, `examples/09_audit_bundle.py`, `examples/10_compliance_overlay.py`, `examples/README.md`

Host AI hard rules:
- Without repo_clone_verified=true, do not claim that the source code has been read.
- Without repo_inspection_verified=true, do not write README, docs, or package-file conclusions as facts.
- Without quick_start_verified=true, do not claim that the Quick Start path has run successfully.

## Doramagic Pitfall Constraints

These rules come from Doramagic discovery, validation, or compilation findings. The host AI must treat them as operating constraints, not background notes.

### Constraint 1: Capability evidence risk requires verification

- Trigger: README/documentation is current enough for a first validation pass.
- Host AI rule: Reproduce the official install and quickstart path in an isolated environment.
- Why it matters: May increase setup, validation, or first-run risk for the user.
- Evidence: capability.assumptions | https://github.com/DanceNitra/inspeximus
- Hard boundary: Do not present this pitfall as solved, verified, or ignorable unless later evidence explicitly closes it.

### Constraint 2: Security or permission risk requires verification

- Trigger: no_demo
- Host AI rule: Reproduce the official install and quickstart path in an isolated environment.
- Why it matters: May increase setup, validation, or first-run risk for the user.
- Evidence: downstream_validation.risk_items | https://github.com/DanceNitra/inspeximus
- Hard boundary: Do not present this pitfall as solved, verified, or ignorable unless later evidence explicitly closes it.

### Constraint 3: Security or permission risk requires verification

- Trigger: no_demo
- Host AI rule: Reproduce the official install and quickstart path in an isolated environment.
- Why it matters: May increase setup, validation, or first-run risk for the user.
- Evidence: risks.scoring_risks | https://github.com/DanceNitra/inspeximus
- Hard boundary: Do not present this pitfall as solved, verified, or ignorable unless later evidence explicitly closes it.

### Constraint 4: Maintenance risk requires verification

- Trigger: issue_or_pr_quality=unknown。
- Host AI rule: Reproduce the official install and quickstart path in an isolated environment.
- Why it matters: May increase setup, validation, or first-run risk for the user.
- Evidence: evidence.maintainer_signals | https://github.com/DanceNitra/inspeximus
- Hard boundary: Do not present this pitfall as solved, verified, or ignorable unless later evidence explicitly closes it.

### Constraint 5: Maintenance risk requires verification

- Trigger: release_recency=unknown。
- Host AI rule: Reproduce the official install and quickstart path in an isolated environment.
- Why it matters: May increase setup, validation, or first-run risk for the user.
- Evidence: evidence.maintainer_signals | https://github.com/DanceNitra/inspeximus
- Hard boundary: Do not present this pitfall as solved, verified, or ignorable unless later evidence explicitly closes it.