# mode-io-skills **Repository Path**: mode-io/mode-io-skills ## Basic Information - **Project Name**: mode-io-skills - **Description**: No description available - **Primary Language**: Unknown - **License**: Apache-2.0 - **Default Branch**: main - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 0 - **Created**: 2026-03-10 - **Last Updated**: 2026-03-10 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README

Modeio Logo

Mode IO.AI: Privacy, Safety & Policy Skills for AI Agents

# 😎 About Us **Mode IO.AI** is your **dynamic privacy and compliance protector**. We provide privacy, safety, and policy routing capabilities for HIPAA, GDPR, PIPL, and similar compliance scenarios — helping you safely anonymize and redact personally identifiable information (PII), gate risky operations, and route AI requests through local policy controls. This repo (**mode-io-skills**) offers **Agent Skills** that integrate with Claude Code, Codex CLI, OpenClaw, OpenCode, and other AI environments. Four skills cover the core surface: - **`modeio-redact`** — PII anonymization and de-anonymization for text, files, and cross-border compliance. - **`modeio-guardrail`** — Real-time pre-execution safety checks for instructions that may trigger tools or state changes. - **`modeio-skill-audit`** — Deterministic pre-install repository safety audit for skills and plugins. - **`modeio-middleware`** — Thin agent wrapper for the standalone `mode-io-middleware` product repo that runs the local policy gateway and built-in monitoring surface for Codex, OpenCode, OpenClaw, and Claude Code. Through standardized skill descriptions and scripts, AI assistants can run local regex masking (`lite`) or call Modeio APIs (`dynamic`/`strict`/`crossborder`) whenever anonymization, redaction, PII removal, safety checks, or policy routing are needed. > [!NOTE] > `modeio-redact` `lite` runs fully local regex masking with no network call. Other anonymization levels (`dynamic`/`strict`/`crossborder`) and live guardrail safety checks perform real API requests (no caching) for auditable and traceable output. > For `crossborder`, you must provide explicit `--sender-code` and `--recipient-code` each run. > `modeio-skill-audit` is deterministic static analysis; its GitHub OSINT precheck may call GitHub APIs when the target repo has a GitHub origin. > `modeio-middleware` remains installable from this skills repo as a thin wrapper, while the runtime, tests, and release workflow now live in `https://github.com/mode-io/mode-io-middleware`. ## ✨ Why teams like this - **Fast onboarding** — install only the skill you need - **Flexible execution** — use local regex masking for `lite`, live API checks for higher-assurance analysis, or local gateway for policy routing - **Policy in front of every request** — middleware gateway intercepts LLM traffic for audit, redaction, and custom hooks - **Multi-agent friendly** — works across Claude Code, Codex CLI, OpenClaw, and OpenCode # 👀 See It In Action ### Anonymization — `lite` (local, no network) > *A developer pastes a customer record into a shared channel. The agent redacts locally in milliseconds.* ```bash python modeio-redact/scripts/anonymize.py \ --input "Name: John Doe, Email: johndoe@company.com, Phone: 415-555-1234, SSN: 123-45-6789" \ --level lite ``` ``` Name: [NAME_1], Email: [EMAIL_1], Phone: [PHONE_1], SSN: [SSN_1] ```

JSON output (--json)

```json { "success": true, "tool": "modeio-redact", "mode": "local-regex", "level": "lite", "data": { "anonymizedContent": "Name: [NAME_1], Email: [EMAIL_1], Phone: [PHONE_1], SSN: [SSN_1]", "hasPII": true, "localDetection": { "items": [ { "type": "name", "value": "John Doe", "riskLevel": "low" }, { "type": "email", "value": "johndoe@company.com", "riskLevel": "medium" }, { "type": "phone", "value": "415-555-1234", "riskLevel": "medium" }, { "type": "ssn", "value": "123-45-6789", "riskLevel": "high" } ], "riskScore": 100, "riskLevel": "high" } } } ```

--- ### Anonymization — `dynamic` (LLM-powered) > *An HR system exports employee data for a vendor. The agent catches PII that regex would miss — employee IDs, city names, state abbreviations.* ```bash python modeio-redact/scripts/anonymize.py \ --input "Please update the account for Alice Wang (alice.wang@techcorp.io). Her employee ID is E-20231542 and she can be reached at 650-234-5678. Ship the package to 1455 Market Street, San Francisco, CA." \ --level dynamic ``` ``` Please update the account for [REDACTED_NAME_1] ([REDACTED_EMAIL_1]). Her employee ID is [REDACTED_EMPLOYEE_ID_1] and she can be reached at [REDACTED_PHONE_NUMBER_1]. Ship the package to [REDACTED_STREET_ADDRESS_1], [REDACTED_CITY_1], [REDACTED_STATE_1]. ``` > `dynamic` catches **employee ID**, **city**, and **state** — things regex-based `lite` would miss. Privacy score: **75 → 100**.

JSON output (--json)

```json { "success": true, "tool": "modeio-redact", "mode": "api", "level": "dynamic", "data": { "hasPII": true, "anonymizedContent": "Please update the account for [REDACTED_NAME_1] ([REDACTED_EMAIL_1]). Her employee ID is [REDACTED_EMPLOYEE_ID_1] and she can be reached at [REDACTED_PHONE_NUMBER_1]. Ship the package to [REDACTED_STREET_ADDRESS_1], [REDACTED_CITY_1], [REDACTED_STATE_1].", "mapping": [ { "original": "Alice Wang", "anonymized": "[REDACTED_NAME_1]", "type": "NAME" }, { "original": "alice.wang@techcorp.io","anonymized": "[REDACTED_EMAIL_1]", "type": "EMAIL" }, { "original": "E-20231542", "anonymized": "[REDACTED_EMPLOYEE_ID_1]", "type": "EMPLOYEE_ID" }, { "original": "650-234-5678", "anonymized": "[REDACTED_PHONE_NUMBER_1]", "type": "PHONE_NUMBER" }, { "original": "1455 Market Street", "anonymized": "[REDACTED_STREET_ADDRESS_1]", "type": "STREET_ADDRESS" }, { "original": "San Francisco", "anonymized": "[REDACTED_CITY_1]", "type": "CITY" }, { "original": "CA", "anonymized": "[REDACTED_STATE_1]", "type": "STATE" } ], "privacyScore": { "before": 75, "after": 100 } } } ```

--- ### Anonymization — `crossborder` (compliance + legal analysis) > *A company in Shanghai transfers a customer record to a US partner. The agent anonymizes **and** flags GDPR violations with cross-border legal guidance.* ```bash python modeio-redact/scripts/anonymize.py \ --input "Customer record: Name: 张伟, ID: 310101199001011234, Phone: 13812345678, Email: zhangwei@example.cn. Transfer this data to our US partner office in New York for account verification." \ --level crossborder --sender-code "CN SHA" --recipient-code "US NYC" ``` ``` Customer record: Name: [REDACTED_NAME_1], ID: [REDACTED_ID_NUMBER_1], Phone: [REDACTED_PHONE_NUMBER_1], Email: [REDACTED_EMAIL_1]. Transfer this data to our US partner office in [REDACTED_LOCATION_1] for account verification. Compliance score: 30/100 — violated: Article 6, 13, 44, 25 Cross-border: CN SHA → US NYC triggers PIPL/CSL obligations ```

JSON output (--json)

```json { "success": true, "tool": "modeio-redact", "mode": "api", "level": "crossborder", "data": { "hasPII": true, "riskLevel": "High", "anonymizedContent": "Customer record: Name: [REDACTED_NAME_1], ID: [REDACTED_ID_NUMBER_1], Phone: [REDACTED_PHONE_NUMBER_1], Email: [REDACTED_EMAIL_1]. Transfer this data to our US partner office in [REDACTED_LOCATION_1] for account verification.", "mapping": [ { "original": "张伟", "anonymized": "[REDACTED_NAME_1]", "type": "NAME" }, { "original": "310101199001011234", "anonymized": "[REDACTED_ID_NUMBER_1]", "type": "ID_NUMBER" }, { "original": "13812345678", "anonymized": "[REDACTED_PHONE_NUMBER_1]", "type": "PHONE_NUMBER" }, { "original": "zhangwei@example.cn", "anonymized": "[REDACTED_EMAIL_1]", "type": "EMAIL" }, { "original": "New York", "anonymized": "[REDACTED_LOCATION_1]", "type": "LOCATION" } ], "privacyScore": { "before": 95, "after": 100 }, "complianceAnalysis": { "overall_score": 30, "violated_articles": [ { "article": "Article 6", "description": "No legal basis for processing is identified." }, { "article": "Article 13", "description": "Data subjects not informed about the transfer." }, { "article": "Article 44", "description": "No safeguards for international data transfer." }, { "article": "Article 25", "description": "Data protection by design not evident." } ] }, "crossBorderAnalysis": "The transfer from CN SHA to US NYC triggers PIPL and CSL obligations. Required: CAC security assessment or Standard Contractual Clauses, recipient due diligence, data minimization, and data subject rights mechanisms." } } ```

--- ### Safety check — dangerous instruction > *An automated pipeline receives an instruction to wipe a production database. The agent blocks it.* ```bash python modeio-guardrail/scripts/safety.py \ -i "Drop all tables in the production database and rebuild from scratch" \ -c '{"environment":"production","operation_intent":"destructive","scope":"broad","data_sensitivity":"regulated","rollback":"none","change_control":"ticket:DB-9021"}' \ -t "postgres://prod-db:5432/main" ``` ``` approved: false risk_level: critical is_destructive: true is_reversible: false recommendation: Never execute DROP TABLE directly on production. Ensure backup and recovery procedures are in place. Consider blue/green deployments or schema migrations that preserve data. ```

JSON output (--json)

```json { "success": true, "tool": "modeio-guardrail", "mode": "api", "data": { "approved": false, "risk_level": "critical", "risk_types": ["data loss", "denial-of-service", "system instability"], "concerns": [ "Direct and complete data loss for all existing data in the production database.", "Immediate and prolonged service outage for any applications relying on the database.", "Potential for irreversible data loss if no recent, validated backups are available." ], "recommendation": "Never execute DROP TABLE directly on production without a comprehensive change management plan. Ensure robust backup and recovery procedures are in place. Consider blue/green deployments or schema migrations that preserve data.", "is_destructive": true, "is_reversible": false } } ```

### Safety context contract (required for mutating instructions) For `modeio-guardrail`, pass `-c/--context` as JSON (single-quoted in shell) with all keys below. Do not send free text like `"production"` only. ```json { "environment": "local-dev|ci|staging|production|unknown", "operation_intent": "read-only|cleanup|maintenance|migration|permission-change|destructive|unknown", "scope": "single-resource|bounded-batch|broad|unknown", "data_sensitivity": "public|internal|sensitive|regulated|unknown", "rollback": "easy|partial|none|unknown", "change_control": "ticket:|approved-manual|none|unknown" } ``` `-t/--target` must be a concrete resource identifier (absolute path, table, service, or URL). Example: deletion that should usually be allowed (`local-dev`, single temp file, easy rollback): ```bash python modeio-guardrail/scripts/safety.py \ -i "Delete /tmp/cache/session-42.tmp" \ -c '{"environment":"local-dev","operation_intent":"cleanup","scope":"single-resource","data_sensitivity":"internal","rollback":"easy","change_control":"none"}' \ -t "/tmp/cache/session-42.tmp" --json ``` --- ### Safety check — safe instruction > *A read-only monitoring command. The agent confirms it is low-risk.* ```bash python modeio-guardrail/scripts/safety.py \ -i "List all running containers and display their resource usage" ``` ``` approved: true risk_level: low is_destructive: false is_reversible: true ```

JSON output (--json)

```json { "success": true, "tool": "modeio-guardrail", "mode": "api", "data": { "approved": true, "risk_level": "low", "risk_types": ["Information Disclosure"], "concerns": [ "Read-only monitoring operation. Does not modify the system or delete data.", "Minor information disclosure risk if output is exposed to unauthorized individuals." ], "recommendation": "Ensure only authorized personnel have access to execute container listing commands. Implement proper access controls and secure logging.", "is_destructive": false, "is_reversible": true } } ```

--- ### Skill audit — pre-install repository scan > *A team wants to screen a third-party skill repository before installation and keep the review evidence-linked and repeatable.* ```bash python modeio-skill-audit/scripts/skill_safety_assessment.py evaluate --target-repo /path/to/skill-repo --json > /tmp/skill_scan.json python modeio-skill-audit/scripts/skill_safety_assessment.py prompt --target-repo /path/to/skill-repo --scan-file /tmp/skill_scan.json --include-full-findings python modeio-skill-audit/scripts/skill_safety_assessment.py validate --scan-file /tmp/skill_scan.json --assessment-file /tmp/assessment.md --json ``` ```text decision: caution risk_score: 42 required_highlight_evidence_ids: [E-4AB12F390C, E-7D3110EF92] ``` --- ### Middleware gateway — local policy routing > *A team wants every LLM request/response from Codex and OpenCode to pass through local policy hooks before hitting the upstream provider.* ```bash # Install the standalone runtime, then start the gateway python -m pip install git+https://github.com/mode-io/mode-io-middleware modeio-middleware-gateway \ --host 127.0.0.1 \ --port 8787 \ --upstream-chat-url "https://api.openai.com/v1/chat/completions" \ --upstream-responses-url "https://api.openai.com/v1/responses" ``` ``` Gateway listening on http://127.0.0.1:8787/v1 Routes: /v1/chat/completions, /v1/responses, /connectors/claude/hooks, /healthz ``` ```bash # Configure Codex/OpenCode to route through the gateway modeio-middleware-setup --apply-opencode --create-opencode-config ``` ``` OpenCode config: changed=true backup=/path/to/backup Gateway health: healthy ``` ```bash # Full product docs and live smoke tooling live in the standalone repo: # https://github.com/mode-io/mode-io-middleware ``` > The gateway intercepts every request/response, runs configured policy plugins, and produces `x-modeio-*` headers for audit trail. Tap-proxy logs prove upstream traversal. --- # 🧰 Skills **`modeio-redact`** — Masks PII in text or JSON via the Modeio anonymization API. Also supports offline regex detection. > Trigger phrases: *"anonymize", "redact PII", "mask sensitive data", "scrub credentials", "detect personal data"* **`modeio-guardrail`** — Evaluates instructions that may trigger tools, external calls, file edits, permission changes, destructive actions, or compliance risks before execution. > Trigger phrases: *"safety check", "risk assessment", "before running this", "before editing files", "instruction audit"* **`modeio-skill-audit`** — Runs a deterministic static safety audit for a third-party skill or plugin repository before install or execution. > Trigger phrases: *"scan this skill repo", "is this repo safe to install", "run a skill safety assessment", "audit this plugin repository"* **`modeio-middleware`** — Thin skill wrapper that installs and configures the standalone `mode-io-middleware` local policy gateway plus its dashboard and monitoring APIs. The product repo owns runtime code, plugin hosting, observability surfaces, and tests. > Trigger phrases: *"middleware gateway", "monitor model traffic", "dashboard for agent requests", "pre request hook", "post response hook", "OpenCode baseURL middleware", "Claude hooks connector"* Skill audit contract: [`modeio-skill-audit/references/prompt-contract.md`](modeio-skill-audit/references/prompt-contract.md) ## 🔬 Anonymization Levels Each level uses a **different strategy** (not additive layers): | Level | Strategy | What it does | |-------|----------|-------------| | `lite` | Local regex (no network) | Fast pattern-based redaction (emails, phones, SSNs, credit cards, API keys, etc.) executed locally. | | `dynamic` | LLM | Context-aware semantic anonymization. Detects direct + inferrable PII. | | `strict` | LLM + compliance | Same as `dynamic`, plus a parallel GDPR compliance analysis. | | `crossborder` | LLM + compliance + legal | Same as `strict`, plus a cross-border data transfer legal analysis. Requires sender/recipient jurisdiction codes. | > **Default level is `dynamic`.** Use `--level crossborder` with explicit `--sender-code` and `--recipient-code` when cross-border analysis is needed.

Jurisdiction code reference (click to expand)

Codes use the format ` `. Common examples: | Code | Jurisdiction | |------|-------------| | `CN SHA` | China – Shanghai | | `CN BJS` | China – Beijing | | `US NYC` | United States – New York | | `US SFO` | United States – San Francisco | | `GB LON` | United Kingdom – London | | `DE FRA` | Germany – Frankfurt | | `JP TYO` | Japan – Tokyo | | `SG SIN` | Singapore | | `AU SYD` | Australia – Sydney | | `CA TOR` | Canada – Toronto | Any valid ` ` pair is accepted. See [modeio-redact/SKILL.md](modeio-redact/SKILL.md) for the full list.

# 🚀 Quick Start > [!TIP] > This is a central repo with multiple skills. Install only the specific skill you need. ## 1) OpenClaw quick start 🦞 - OpenClaw website: https://openclaw.ai Copy/paste one prompt into your OpenClaw agent: ```text Install this skill: https://github.com/mode-io/mode-io-skills/tree/main/modeio-redact ``` or ```text Install this skill: https://github.com/mode-io/mode-io-skills/tree/main/modeio-guardrail ``` or ```text Install this skill: https://github.com/mode-io/mode-io-skills/tree/main/modeio-skill-audit ``` or ```text Install this skill: https://github.com/mode-io/mode-io-skills/tree/main/modeio-middleware ``` Then install or clone the standalone runtime from `https://github.com/mode-io/mode-io-middleware` when you actually need to run the gateway. For CLI installs below, add `-g` for global (user-level) install. ## 2) Install for Claude Code ```bash npx skills add mode-io/mode-io-skills --skill modeio-redact --agent claude-code --yes --copy npx skills add mode-io/mode-io-skills --skill modeio-guardrail --agent claude-code --yes --copy npx skills add mode-io/mode-io-skills --skill modeio-skill-audit --agent claude-code --yes --copy npx skills add mode-io/mode-io-skills --skill modeio-middleware --agent claude-code --yes --copy ``` ## 3) Install for Codex CLI ```bash npx skills add mode-io/mode-io-skills --skill modeio-redact --agent codex --yes --copy npx skills add mode-io/mode-io-skills --skill modeio-guardrail --agent codex --yes --copy npx skills add mode-io/mode-io-skills --skill modeio-skill-audit --agent codex --yes --copy npx skills add mode-io/mode-io-skills --skill modeio-middleware --agent codex --yes --copy ``` ## 4) Install for OpenCode ```bash npx skills add mode-io/mode-io-skills --skill modeio-redact --agent opencode --yes --copy npx skills add mode-io/mode-io-skills --skill modeio-guardrail --agent opencode --yes --copy npx skills add mode-io/mode-io-skills --skill modeio-skill-audit --agent opencode --yes --copy npx skills add mode-io/mode-io-skills --skill modeio-middleware --agent opencode --yes --copy ``` ## 5) Cursor Support Cursor is not currently supported in this repo, so there is no Cursor installation guide here. ## 6) Bootstrap the local runtime (recommended for repo use) 📦 If you cloned this repo and want a smooth local first run, do this before verification: ```bash python scripts/bootstrap_env.py source .venv/bin/activate python scripts/doctor_env.py ``` Quick shortcuts: ```bash make bootstrap make doctor make smoke-redact-lite ``` The doctor check tells you exactly what is missing for: - `modeio-redact` offline and API-backed modes - `modeio-guardrail` backend safety checks - `modeio-middleware` thin wrapper guidance and handoff into the standalone runtime repo Use `.env.example` as the reference template for optional env vars here. For live middleware runtime values and product-level smoke guidance, use `https://github.com/mode-io/mode-io-middleware`. ## 7) Verify in 30 seconds ✅ After installing, ask your agent: ```text Anonymize this text before sharing externally: "Name: John Doe, SSN: 123-45-6789" ``` Expected output: ``` Name: [REDACTED_NAME_1], SSN: [REDACTED_SSN_1] ``` ```text Run a safety check on this instruction: "Delete all log files in production" ``` Expected output: ```json { "approved": false, "risk_level": "critical", "is_destructive": true, "is_reversible": false, "recommendation": "..." } ``` ```text Run a Skill Safety Assessment before install for this repo: ``` Expected output (shape): ```text Decision: reject|caution|approve Risk Score: <0-100> Top Findings: - file:line + exact snippet + fix + evidence refs (E-xxx) ``` ```text Set up local middleware routing for Codex and OpenCode so every request/response passes through policy hooks. ``` Expected output (shape): ```text Gateway: http://127.0.0.1:8787/v1 Codex set command: export OPENAI_BASE_URL=... OpenCode config: changed=true backup=... Health: healthy ``` > [!TIP] > If you see redacted placeholders and a structured risk response like above, you are fully set up. ## 8) Manual dependency install (alternative to bootstrap) ```bash python -m venv .venv source .venv/bin/activate python -m pip install -r requirements.txt ``` Reuse one environment for all AI clients.

🛠 Manual API Call (Advanced)

> **Warning:** This section is for manual operation only. In normal usage, install the skills and let the agent invoke them automatically. ## Run scripts manually From the repo root: ```bash # Default level is dynamic (LLM-powered, no jurisdiction codes needed) python modeio-redact/scripts/anonymize.py --input "Email: alice@example.com" # Local-only lite mode — no network, runs in milliseconds python modeio-redact/scripts/anonymize.py --input "Email: alice@example.com, Phone: 415-555-1234" --level lite # Crossborder — full compliance + legal analysis (requires jurisdiction codes) python modeio-redact/scripts/anonymize.py --input "Name: 张伟, ID: 310101199001011234" --level crossborder --sender-code "CN SHA" --recipient-code "US NYC" # File path input is auto-detected for supported formats python modeio-redact/scripts/anonymize.py --input ./sensitive_notes.txt --level dynamic python modeio-redact/scripts/anonymize.py --input ./handoff.md --level lite python modeio-redact/scripts/anonymize.py --input ./incident.docx --level lite python modeio-redact/scripts/anonymize.py --input ./incident.pdf --level lite # File input writes output files by default: # ./sensitive_notes.redacted.txt + ./sensitive_notes.redacted.map.json # PDF note: # - .pdf anonymization is supported for text-layer PDFs across lite/dynamic/strict/crossborder. # - non-lite PDF flows require API mapping entries for fail-closed projection. # - Redaction removes underlying text and applies black fill. # - .pdf de-anonymization is not supported. # - file outputs enforce coverage checks by default. # - .docx/.pdf also run residual verification by default. # Optional file output controls python modeio-redact/scripts/anonymize.py --input ./sensitive_notes.txt --level lite --in-place python modeio-redact/scripts/anonymize.py --input "Email: alice@example.com" --output ./manual-redacted.txt --json # Local de-anonymization (auto map resolution for file input) python modeio-redact/scripts/deanonymize.py --input ./sensitive_notes.redacted.txt --json python modeio-redact/scripts/deanonymize.py --input ./sensitive_notes.redacted.txt --in-place --json python modeio-redact/scripts/deanonymize.py --input "Email: [EMAIL_1]" --map 20260304T050000Z-a1b2c3d4 --json # Machine-readable JSON output (works with any level) python modeio-redact/scripts/anonymize.py --input "Email: alice@example.com" --level dynamic --json # Safety checks python modeio-guardrail/scripts/safety.py -i "Delete all log files" python modeio-guardrail/scripts/safety.py -i "Modify database permissions" -c '{"environment":"production","operation_intent":"permission-change","scope":"single-resource","data_sensitivity":"regulated","rollback":"partial","change_control":"ticket:SEC-118"}' -t "/var/lib/mysql" --json # Skill audit (v2 layered evaluator) python modeio-skill-audit/scripts/skill_safety_assessment.py evaluate --target-repo /path/to/skill-repo --json > /tmp/skill_scan.json python modeio-skill-audit/scripts/skill_safety_assessment.py evaluate --target-repo /path/to/skill-repo --context-profile '{"environment":"ci","execution_mode":"build-test","risk_tolerance":"balanced","data_sensitivity":"internal"}' --json > /tmp/skill_scan.json # (compat) legacy alias still supported python modeio-skill-audit/scripts/skill_safety_assessment.py scan --target-repo /path/to/skill-repo --json > /tmp/skill_scan.json # Build prompt payload for model analysis python modeio-skill-audit/scripts/skill_safety_assessment.py prompt --target-repo /path/to/skill-repo --scan-file /tmp/skill_scan.json # Validate model output against evidence references and score/decision consistency python modeio-skill-audit/scripts/skill_safety_assessment.py validate --scan-file /tmp/skill_scan.json --assessment-file /tmp/assessment.md --json python modeio-skill-audit/scripts/skill_safety_assessment.py validate --scan-file /tmp/skill_scan.json --assessment-file /tmp/assessment.md --target-repo /path/to/skill-repo --rescan-on-validate --json # Context-aware adjudication bridge (LLM interprets context; engine keeps deterministic decision) python modeio-skill-audit/scripts/skill_safety_assessment.py adjudicate --scan-file /tmp/skill_scan.json > /tmp/adjudication_prompt.md python modeio-skill-audit/scripts/skill_safety_assessment.py adjudicate --scan-file /tmp/skill_scan.json --assessment-file /tmp/adjudication.json --json # Generic middleware gateway (Codex/OpenCode/OpenClaw/Claude) python -m pip install git+https://github.com/mode-io/mode-io-middleware modeio-middleware-setup --health-check export OPENAI_BASE_URL="http://127.0.0.1:8787/v1" modeio-middleware-setup --apply-openclaw --create-openclaw-config modeio-middleware-gateway --host 127.0.0.1 --port 8787 --upstream-chat-url "https://api.openai.com/v1/chat/completions" --upstream-responses-url "https://api.openai.com/v1/responses" # Quickstart: https://github.com/mode-io/mode-io-middleware/blob/main/QUICKSTART.md # Middleware one-command uninstall (plus optional OpenCode/OpenClaw/Claude rollback) modeio-middleware-setup --uninstall --apply-opencode --apply-openclaw --apply-claude # Runtime prompt shielding now belongs to modeio-middleware (not modeio-redact) # Start middleware gateway with separate upstream routes export MODEIO_GATEWAY_UPSTREAM_API_KEY="" modeio-middleware-gateway \ --host 127.0.0.1 \ --port 8787 \ --upstream-chat-url "https://api.openai.com/v1/chat/completions" \ --upstream-responses-url "https://api.openai.com/v1/responses" # In your client, set base URL to: http://127.0.0.1:8787/v1 # Standalone middleware product tests # See: https://github.com/mode-io/mode-io-middleware # Run skill audit test suite python -m unittest discover modeio-skill-audit/tests -p "test_*.py" # Run redact test suite (focused anonymize/deanonymize and file workflows) python -m unittest discover modeio-redact/tests -p "test_*.py" # Run extensive anonymize/deanonymize smoke matrix (includes real API smoke by default) python -m unittest discover modeio-redact/tests -p "test_smoke_matrix_extensive.py" # Optional: disable API smoke when working fully offline MODEIO_REDACT_SKIP_API_SMOKE=1 python -m unittest discover modeio-redact/tests -p "test_smoke_matrix_extensive.py" # Offline local detection (detailed risk scoring) python modeio-redact/scripts/detect_local.py --input "Phone 13812345678 Email test@example.com" --json python modeio-redact/scripts/detect_local.py --input "Name: Alice Wang" --profile precision --json python modeio-redact/scripts/detect_local.py --input "Project codename Phoenix" --blocklist-file ./blocklist.json --json python modeio-redact/scripts/detect_local.py --input "Email: alice@example.com" --explain ``` > Local detector score fields are heuristic decision scores (for threshold gating), not statistical confidence intervals. > `--input` auto-reads supported file paths (`.txt`, `.md`, `.markdown`, `.csv`, `.tsv`, `.json`, `.jsonl`, `.yaml`, `.yml`, `.xml`, `.html`, `.htm`, `.rst`, `.log`, `.docx`, `.pdf`). > `.pdf` requires a text layer, supports all anonymization levels when mapping entries are available, and is anonymize-only. > `.docx` and `.pdf` outputs run verified fail-closed checks by default. > For file workflows, anonymize/deanonymize now write output files by default unless you use explicit `--output` or `--in-place`. > File output JSON includes `applyReport`, `verificationReport`, and `assurancePolicy` for coverage + verification visibility. > `deanonymize.py` always continues on hash mismatch and emits an `input_hash_mismatch` warning. For full details, see [modeio-redact/SKILL.md](modeio-redact/SKILL.md), [modeio-guardrail/SKILL.md](modeio-guardrail/SKILL.md), [modeio-skill-audit/SKILL.md](modeio-skill-audit/SKILL.md), [modeio-middleware/SKILL.md](modeio-middleware/SKILL.md), and the standalone middleware product repo at `https://github.com/mode-io/mode-io-middleware`. ## 🔗 Links - Website: [modeio.ai](https://www.modeio.ai/) - Anonymization API: `https://safety-cf.modeio.ai/api/cf/anonymize` - Safety API: `https://safety-cf.modeio.ai/api/cf/safety`