[vchamarthi]

Summary

dokimi repo fetch prints <sha>\t<git_dir>. A script has to split a tab-separated line to get the pinned commit. --json prints {"sha", "git_dir"} instead, the same flag repo validate, scan and inspect already take. The default output is unchanged.

Evidence

• base: 5c70cbf90f4eab95b30dcd6ad7ae60a4f37de268
• command: run.sh under experiment_safe.py (macOS sandbox): python3.14 -m pytest tests/test_repo_manifest.py -k repo_fetch_cli, and tests/test_docs_code_refs.py
• before: exit 2. test_repo_fetch_cli_json failed with dokimi: error: unrecognized arguments: --json.
• after: test_repo_fetch_cli_json passes. It checks the JSON keys and that the default line is still <sha>\t<git_dir>. tests/test_docs_code_refs.py: 31 passed.
• not claimed: the existing git-backed test_repo_fetch_cli could not run in the sandbox (git cannot open /dev/null there), so it was not re-run. The new test stubs open_source, so it does not clone. make ci-local was not run. No live fetch.

Blast radius

cmd_repo_fetch and the fetch parser in dokimi/dokimi/commands/repo.py, one new test, one line in repo-evaluation-contract.md, and one changelog fragment. Without --json the output is byte-for-byte the same.

Overlap

• checked: intel/dokimi
• intel/dokimi open: 24
• vchamarthi/dokimi open: 5
• fix(intents): a domain with a blank products or name is listed #71 not the same: it adds repo budget. It does not touch repo fetch.
• ci: reusable repo-eval workflow (validate, scan, tasks check) with dokimi pinned intel/dokimi#113 not the same: it adds a reusable repo-eval workflow. It does not touch repo fetch.
• feat(repos): built-in review graders dokimi:review_seeds and dokimi:review_replay intel/dokimi#120 not the same: it adds built-in review graders. It does not touch repo fetch.
• feat(repo): record the model that answered each cell intel/dokimi#119 not the same: it records the model that answered each cell. It does not touch repo fetch.
• feat(repo): Codex and Copilot CLI harness profiles, native review mode intel/dokimi#116 not the same: it adds Codex and Copilot harness profiles. It does not touch repo fetch.
• repo scan: L0 rules for parity with oneDAL static_check (D6) intel/dokimi#115 not the same: it adds L0 rules. It does not touch repo fetch.
• feat(repo): compare flags comparisons with no headroom intel/dokimi#114 not the same: it flags comparisons with no headroom. It does not touch repo fetch.
• feat(repo): task host requirements, unsupported instead of a false pass (#41 D3) intel/dokimi#112 not the same: it adds task host requirements. It does not touch repo fetch.
• feat(repo): --delegation and an execution id for repo run cells intel/dokimi#107 not the same: it adds --delegation and an execution id. It does not touch repo fetch.
• ci(docs): flag docs pages whose depends_on code a PR changes intel/dokimi#92 not the same: it flags docs pages whose code changed. It does not touch repo fetch.

Created with claude-sonnet-5-5.

[vchamarthi]

Superseded by intel#149 and intel#150.