fix(gh): rewrite gh --json/--jq/--template instead of deferring

[dopaemon]

Fixes #4067

Problem

$ rtk rewrite 'gh pr view 123 --json reviews'
$ echo $?
1

Exit 1 / empty output is the "no rewrite" verdict, so the hook passes the command through untouched. Since almost every real gh pr view call carries --json, that whole family was invisible to rtk — the issue reports it as the single largest missed-savings command in rtk discover.

Why the guard is stale

registry.rs skipped the rewrite for rtk gh whenever the args contained --json/--jq/--template (#196), on the grounds that rtk gh would corrupt structured output. That hasn't been true for a while: gh_cmd::run checks the same flags first and hands the call to run_passthrough, which streams gh's own bytes and propagates its exit code — the same stance gh api documents ("Passthrough preserves the full response and tracks metrics at 0% savings").

So the guard bought nothing and cost coverage.

Changes

• src/discover/registry.rs: drop the rtk gh + --json/--jq/--template special-case. gh pr view 123 --json reviews now rewrites to rtk gh pr view 123 --json reviews.

• src/cmds/git/gh_cmd.rs: harden the passthrough gate the registry guard was masking. has_json_flag matched only a bare --json, so the = spelling slipped into the filter path. On current develop:

  $ rtk gh pr view 4066 --repo rtk-ai/rtk --json=number
  [open] PR #4066: fix(read): fall back to raw bytes ...   # rtk's summary, not the requested field

  wants_raw_output replaces it and covers --json/--jq/-q/--template/-t in both --flag value and --flag=value forms. Without this, lifting the registry guard would have made that wrong-output path reachable from the hook.

• Tests: the four fix: passthrough --json/--jq/--template flags in all gh subcommands #196 registry tests now assert the rewrite instead of None (plus one for the issue's exact gh pr view 123 --json reviews), and three unit tests cover the flag spellings.

Verification

Byte-comparison against a real PR, gh vs rtk gh:

$ gh pr view 4066 --repo rtk-ai/rtk --json number,title,state,reviews > a.json
$ rtk gh pr view 4066 --repo rtk-ai/rtk --json number,title,state,reviews > b.json
$ cmp a.json b.json && echo identical
identical                       # 3025 bytes, exit 0 both

$ # --jq and --template forms: identical as well
$ rtk rewrite 'gh pr view 123 --json reviews'
rtk gh pr view 123 --json reviews

cargo fmt --all + cargo clippy --all-targets clean; cargo test --bin rtk → 3615 passed, plus the hook/gh integration binaries (hook_decision_protocol_test, gh_pr_checks_failure_test, copilot_selfheal_test, hook_warning_scope_test) green.

Not included

The issue's secondary note (git log -1 | grep -n foo leaves the post-pipe stage unrewritten) is a separate, deliberate mechanism — PipelineSafety only rewrites stages a filter can safely occupy. Worth its own issue rather than folding into this fix.

[rtk-wshm-sync-bot]

wshm · Automated triage by AI

📊 Automated PR Analysis

🐛 Type	bug-fix
🟡 Risk	medium

Summary

Removes a stale registry guard that skipped rewriting gh commands containing --json/--jq/--template, since rtk gh already passthroughs those flags unchanged via run_passthrough. Also hardens the passthrough flag detection (renamed has_json_flag to wants_raw_output) to catch --json=, --jq, -q, --template, and -t in both space and equals forms, fixing a case where --json=field wasn't detected and rtk's own summary was shown instead of raw gh output.

Review Checklist

• Tests present
• Breaking change
• Docs updated

Linked issues: #4067, #196

---

Analyzed automatically by wshm · This is an automated analysis, not a human review.