# AI use cases for Codex Privacy Hud (inin-zou/codex-privacy-hud) Repo: [inin-zou/codex-privacy-hud](https://github.com/inin-zou/codex-privacy-hud) | Use cases: 3 | Page: https://dedupedai.com/use-cases/inin-zou-codex-privacy-hud/ A local-first Codex plugin that records hook observations in a session ledger, shows a privacy HUD for each session, and can deny or rewrite tool input before it runs. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - See what a Codex session discloses, live: https://dedupedai.com/use-cases/inin-zou-codex-privacy-hud/see-privacy-disclosures-in-codex-session/ - Repair the HUD runtime after a plugin update: https://dedupedai.com/use-cases/inin-zou-codex-privacy-hud/repair-privacy-hud-after-plugin-update/ - Block network commands that name sensitive paths: https://dedupedai.com/use-cases/inin-zou-codex-privacy-hud/block-network-calls-touching-sensitive-files/ ## How to see what a Codex session discloses, live with Codex Privacy Hud Repo: [inin-zou/codex-privacy-hud](https://github.com/inin-zou/codex-privacy-hud) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Install the HUD and read confirmed points, recipients and denials for each Codex session. **Who it is for:** A developer running Codex CLI on a Mac who wants to know whether their agent sent personal data or secrets out during a session. **The problem:** Agent sessions call tools and send context to models, and there is no visible record of what was disclosed. **The outcome:** A per-session HUD that shows confirmed points, distinct disclosures, confirmed recipients and denials issued, with a percentage only when the recorded evidence supports it. ### Steps 1. Run `codex --version` and confirm it prints a version such as codex-cli 0.154.0. 2. Run `python3 --version` and confirm Python is 3.11 or newer. 3. Run `curl -fsSL https://raw.githubusercontent.com/inin-zou/codex-privacy-hud/main/install.sh | sh` to install the plugin. 4. Start a new Codex session and open the Privacy HUD readings for it. 5. Check that the reading shows confirmed points and unresolved actions as separate items, and that a missing percentage is shown as unavailable rather than 0%. ### Prompt to paste into your agent ```text Using the repo inin-zou/codex-privacy-hud, install the plugin on this macOS machine and open the Privacy HUD for the session . Report confirmed points, distinct disclosures, confirmed recipients, denials issued and whether a percentage is shown. Verify the session began from a genuine SessionStart before treating its accounting as evidence-based. ``` ### Caveat Detection is heuristic, sessions are not monitored while the model loads, and hosted tools bypass hooks, so a zero reading does not mean no disclosure occurred. Risk, remote script: The one-line installer downloads and runs a script from GitHub, and the README notes that the checksum only detects corrupted downloads, not a compromised publisher. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/inin-zou-codex-privacy-hud/see-privacy-disclosures-in-codex-session/ ## How to repair the HUD runtime after a plugin update with Codex Privacy Hud Repo: [inin-zou/codex-privacy-hud](https://github.com/inin-zou/codex-privacy-hud) | Difficulty: Advanced | Time: 30 minutes | Basis: From the README Check runtime alignment with doctor, then run the printed repair command in another terminal. **Who it is for:** A Codex user who updated the Privacy HUD plugin and now sees unverified warnings or denials on outbound calls. **The problem:** After an update, old daemons or MCP servers can keep running while the new hooks no longer match the recorded runtime selection. **The outcome:** The runtime is aligned again, the Privacy HUD reports verified ingress, and the Codex app or CLI is restarted with the plugin loaded. ### Steps 1. Run `python3 "$PRIVACY_HUD_BUNDLE/scripts/runtime.py" --plugin-data "$PRIVACY_HUD_DATA" doctor` to check alignment. (inferred, check the README) 2. Run `python3 "$PRIVACY_HUD_BUNDLE/scripts/runtime.py" --plugin-data "$PRIVACY_HUD_DATA" repair --print-command` to print the exact recovery command. (inferred, check the README) 3. Run the printed repair command in another terminal, because repair does not run itself; it may download dependencies and model weights. 4. Restart the Codex app, CLI or IDE integration that loaded the plugin. 5. Run the doctor command again and confirm the runtime reports as aligned. ### Prompt to paste into your agent ```text In the repo inin-zou/codex-privacy-hud, set PRIVACY_HUD_BUNDLE to and PRIVACY_HUD_DATA to , then run the doctor command. Expected output: a report of runtime alignment, and if it is misaligned, the exact repair command. Verify that you do not run the repair command yourself unless the user has approved the downloads it may perform. ``` ### Caveat Repair preserves ledger records but cannot reconstruct monitoring gaps or lost in-memory state from the period before it ran. Risk, remote script: The repair command may download dependencies and model weights, so review the printed command before running it. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/inin-zou-codex-privacy-hud/repair-privacy-hud-after-plugin-update/ ## How to block network commands that name sensitive paths with Codex Privacy Hud Repo: [inin-zou/codex-privacy-hud](https://github.com/inin-zou/codex-privacy-hud) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Use the default-on lexical network guard to deny Codex network commands that reference sensitive files. **Who it is for:** A developer who lets Codex run shell commands on a machine with credential or key files, and wants a denial before data leaves. **The problem:** A network command such as a curl call can send the contents of a sensitive file to a remote host without any visible check. **The outcome:** Recognized network commands that contain known-sensitive-path references are denied, and the denial is counted in the session HUD. ### Steps 1. Run the one-line installer with `curl -fsSL https://raw.githubusercontent.com/inin-zou/codex-privacy-hud/main/install.sh | sh`. 2. Restart Codex so the hooks load from the installed plugin. 3. In a new session, ask Codex to run a network command that references a known-sensitive path in a test directory. 4. Confirm the command is denied and that the denial appears in the Privacy HUD for that session. 5. Check that the denial is recorded separately from the local read guard, which is an independent check. ### Prompt to paste into your agent ```text Using the repo inin-zou/codex-privacy-hud, install the plugin and run a test network command that references inside a Codex session. Expected output: a denial from the lexical network guard and a denials-issued count in the HUD. Verify that no file contents were opened or rewritten by the guard, as the README states. ``` ### Caveat The guard is lexical, so it only catches recognized network commands with known-sensitive-path references, and shell-derived file identities remain unresolved. Risk, remote script: The installer downloads and runs a script from GitHub, so verify the source before piping it to sh. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/inin-zou-codex-privacy-hud/block-network-calls-touching-sensitive-files/