Add OrcaReplay to Agent skills and workflows

[xizhuomengcontin]

Adds OrcaReplay to Agentic software engineering → Agent skills and workflows.

Pillar 3 case. Your curation policy puts "coding-agent harnesses, agent skills, software factories, orchestration, evaluation, and workflows that improve software delivery" in scope. OrcaReplay sits on the evaluation-and-diagnosis side of that: it records a coding-agent run below the harness, at the process and socket boundary, so the model traffic, the shell commands with their exit codes, the per-turn file changes and the MCP JSON-RPC calls land on one timeline. The recording then replays offline with the network off — reproducing the run byte-for-byte on another machine — or forks from a chosen checkpoint onto a different model, with earlier turns still served from the recording so two models face identical context and file state.

The distinct engineering claim, since the policy asks for distinctness rather than coverage: it does not need the harness to cooperate. A Codex CLI signed in with a ChatGPT subscription talks only to its own backend and exposes no base URL to redirect; capture happens a level below that and still works. This is what makes the offline replay possible at all, and it is the part no SDK-based tracing tool does.

It also ships the thing this subsection is actually about: a skill (skills/orca-replay/SKILL.md) that tells an agent to answer questions about a past run from the trace rather than reconstructing events from context, and to keep what the recording shows separate from what it is inferring — including the honest limit that replay proves the recorded decisions still reproduce but cannot prove a fresh run would fail the same way, since the model is not asked again.

On your out-of-scope clause for "personal projects submitted mainly for promotion": I am the author, so that clause plainly applies to how this arrived, and you should weigh it. What I can offer against it: Apache-2.0, 12 packages on npm with sigstore provenance, a validation write-up of a real bug fix recorded, replayed offline end to end and forked, and the same skill reviewed and merged into sickn33/agentic-awesome-skills after fourteen rounds of adversarial review and published to ClawHub past its security scan. The counterweight you should know without asking: the repository is nine days old with 150 stars. If your bar is a track record, this does not have one yet, and holding it for that reason is a decision I would respect rather than argue with.

[greptile-apps]

Greptile Summary

The PR adds OrcaReplay to the “Agent skills and workflows” section of the README.

• Links to the OrcaReplay repository.
• Describes its coding-agent recording, offline replay, checkpoint forking, and trace-reading skill.

Confidence Score: 5/5

The documentation-only change appears safe to merge.

The added resource fits the target subsection, follows the repository’s entry structure, and introduces no concrete correctness, security, or rule-compliance issue.

Important Files Changed

Filename	Overview
README.md	Adds one correctly formatted resource entry under “Agent skills and workflows”; no actionable issue was identified.

_{Reviews (1): Last reviewed commit: "Add OrcaReplay to Agent skills and workf..." | Re-trigger Greptile}