# AI use cases for Orca Replay (Continuum-AI-Corp/OrcaReplay) Repo: [Continuum-AI-Corp/OrcaReplay](https://github.com/Continuum-AI-Corp/OrcaReplay) | Use cases: 5 | Page: https://dedupedai.com/use-cases/continuum-ai-corp-orcareplay/ OrcaReplay is a TypeScript npm tool that records an AI agent run through a local proxy and lets you replay it offline, fork it onto other models, and inspect what happened. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Find the exact step that broke an agent run: https://dedupedai.com/use-cases/continuum-ai-corp-orcareplay/find-the-step-that-broke-an-agent-run/ - Fork a run at one step and compare models: https://dedupedai.com/use-cases/continuum-ai-corp-orcareplay/fork-run-to-compare-models-at-one-step/ - Try Orca Replay with a sample run and no key: https://dedupedai.com/use-cases/continuum-ai-corp-orcareplay/try-orcareplay-without-an-agent-or-key/ - Capture an agent's system prompt for each model: https://dedupedai.com/use-cases/continuum-ai-corp-orcareplay/capture-agent-system-prompt-by-model/ - Record a Codex CLI run that cannot be redirected: https://dedupedai.com/use-cases/continuum-ai-corp-orcareplay/record-codex-cli-that-cannot-be-redirected/ ## How to find the exact step that broke an agent run with Orca Replay Repo: [Continuum-AI-Corp/OrcaReplay](https://github.com/Continuum-AI-Corp/OrcaReplay) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Replay a failed coding agent run offline and see which tool call changed a file or skipped a failing check. **Who it is for:** A developer whose coding agent reported success, but the tests still fail or a file was changed wrongly. **The problem:** The agent's exit code and its transcript do not show which action caused the damage or whether a check actually failed. **The outcome:** A step-by-step timeline and causal graph that shows which tool call changed which file and which check failed, with the replay blocked from calling any model. ### Steps 1. Run `npm i -g orcareplay` to install the package that provides the `orca` command. 2. Run `orca record claude` and use your agent as normal until the run finishes. 3. Run `orca show last` to read the ordered timeline of model turns, tool calls, file changes and shell results. 4. Run `orca graph last` to see which events produced which, and check whether each link is recorded or inferred. 5. Run `orca replay last` to repeat the run with no network and no tokens spent, and confirm the same result. ### Prompt to paste into your agent ```text Use the OrcaReplay repo (Continuum-AI-Corp/OrcaReplay) to inspect the most recent failed run on this machine by running `orca show last` and `orca graph last`. Report the sequence number of the tool call that changed and the sequence number of any shell check that exited non-zero. Verify that `orca replay last` prints the run again with no model called. ``` ### Caveat Replay is not a sandbox, and a tool call that opens its own socket, such as a shell command running curl, is outside the no-network guarantee. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/continuum-ai-corp-orcareplay/find-the-step-that-broke-an-agent-run/ ## How to fork a run at one step and compare models with Orca Replay Repo: [Continuum-AI-Corp/OrcaReplay](https://github.com/Continuum-AI-Corp/OrcaReplay) | Difficulty: Intermediate | Time: 60 minutes | Basis: From the README Keep the first steps of a recorded run fixed, switch to another model from a chosen checkpoint, and compare results. **Who it is for:** A team lead or developer choosing which model to use for a coding task who wants a fair comparison. **The problem:** Comparing models on a task usually changes the starting context too, so results do not show the effect of the model alone. **The outcome:** Replays that match the recording up to checkpoint 4, then continue on a different model, and a comparison table where the model is the only variable. ### Steps 1. Run `npm i -g orcareplay` and record a run with `orca record claude`. 2. Run `orca replay last --from 4 --model claude-haiku-4-5 --ui` to continue the same conversation on a different model from step 4. 3. Run `orca compare last --from 4 --models a,b` with the model names you want to test in place of a and b. 4. Run `orca show last` after each fork to check which turns came from the trace and which came from the network. ### Prompt to paste into your agent ```text Use the OrcaReplay repo (Continuum-AI-Corp/OrcaReplay) to fork the recorded run `last` at checkpoint 4 onto the models and using `orca compare last --from 4 --models ,`. Report one comparison table with the outcome for each model. Verify that turns before checkpoint 4 are identical across both forks. ``` ### Caveat Forked turns after the cursor call live model APIs, so they need keys and network access and they spend tokens. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/continuum-ai-corp-orcareplay/fork-run-to-compare-models-at-one-step/ ## How to try Orca Replay with a sample run and no key with Orca Replay Repo: [Continuum-AI-Corp/OrcaReplay](https://github.com/Continuum-AI-Corp/OrcaReplay) | Difficulty: Beginner | Time: 15 minutes | Basis: From the README Run a built-in demo that replays a recorded agent fix without an agent, a key or any model call. **Who it is for:** A developer evaluating Orca Replay who has no coding agent installed and no API key yet. **The problem:** Setting up an agent, a key and tokens before seeing the tool is a high barrier to deciding whether it is useful. **The outcome:** A small project with a real bug, a recording of an agent fixing it, and a replay showing failing tests before and passing tests after. ### Steps 1. Run `npm i -g orcareplay` to install the package. 2. Run `orca quickstart` to create the sample project and its recording. 3. Run `orca quickstart --full` to print the whole timeline and the replay as it happened. (inferred, check the README) 4. Check that the output shows two failing tests before the fix, four passing tests after, and three turns served from the trace. ### Prompt to paste into your agent ```text Use the OrcaReplay repo (Continuum-AI-Corp/OrcaReplay) by running `orca quickstart` in an empty directory. Report the number of failing tests before the replay and passing tests after it. Verify that the output states no model was called and no tokens were spent. ``` ### Caveat The quickstart only shows the replay of a bundled recording, so it does not test recording your own agent. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/continuum-ai-corp-orcareplay/try-orcareplay-without-an-agent-or-key/ ## How to capture an agent's system prompt for each model with Orca Replay Repo: [Continuum-AI-Corp/OrcaReplay](https://github.com/Continuum-AI-Corp/OrcaReplay) | Difficulty: Intermediate | Time: 30 minutes | Basis: From the README Record the prompt a harness assembles for a given model, with the local machine details removed. **Who it is for:** An engineer comparing how a coding agent behaves on different models who needs the exact prompt used for each. **The problem:** The prompt an agent harness builds is hidden, and it can differ between interactive and print modes and between models. **The outcome:** A captured, sanitised system prompt filed under the model name, which you can compare across models. ### Steps 1. Run `npm i -g orcareplay` and clone the repo if you need the capture folder locally. 2. Run `node capture/capture.mjs claude --model claude-opus-5` to capture the prompt for that model. 3. Read `capture/README.md` for the measured differences between interactive and print prompts and the sanitising rules. 4. Repeat the capture command with another model name to get a second prompt file for comparison. ### Prompt to paste into your agent ```text Use the OrcaReplay repo (Continuum-AI-Corp/OrcaReplay) to capture the system prompt for the agent claude with the model by running `node capture/capture.mjs claude --model `. Report the file path where the sanitised prompt was saved. Verify that the saved prompt contains no local machine paths. ``` ### Caveat The capture script is in the repo under capture/, so it needs a clone of the repo, and the README does not describe its output location beyond filing by model. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/continuum-ai-corp-orcareplay/capture-agent-system-prompt-by-model/ ## How to record a Codex CLI run that cannot be redirected with Orca Replay Repo: [Continuum-AI-Corp/OrcaReplay](https://github.com/Continuum-AI-Corp/OrcaReplay) | Difficulty: Advanced | Time: 60 minutes | Basis: From the README Use opt-in TLS interception to capture a Codex CLI session that talks to its own backend. **Who it is for:** A developer using Codex CLI signed in with a ChatGPT subscription who needs a replayable record of a run. **The problem:** A Codex CLI session uses a credential that only its own backend accepts, so a base-URL redirect cannot capture it. **The outcome:** A trace of the run captured at the process and socket boundary, which you can then replay and inspect with the same commands as other runs. ### Steps 1. Run `npm i -g orcareplay` to install the package. 2. Read the README section on when the harness will not be redirected and confirm your Codex CLI setup matches it. 3. Run `orca record` with the `--tls-intercept` option, following the README for the exact syntax of that flag. (inferred, check the README) 4. Run `orca show last` to confirm the trace includes model turns, and `orca replay last` to confirm the offline replay. ### Prompt to paste into your agent ```text Use the OrcaReplay repo (Continuum-AI-Corp/OrcaReplay) to record a Codex CLI session with TLS interception enabled, following the README section on when the harness will not be redirected. Report whether `orca show last` lists model turns from the session. Verify that hosts not on the intercept list are listed in the trace as tunnelled. ``` ### Caveat The README describes --tls-intercept only at a high level, so the exact flag syntax and setup must be read from the README before running this. Risk, reverse engineering: TLS interception decrypts traffic between a vendor client and its backend, so check that your use complies with the vendor's terms and your organisation's policy before recording. Dual-use notice: only use this on software, sites and systems you own or are authorized to test. 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/continuum-ai-corp-orcareplay/record-codex-cli-that-cannot-be-redirected/