fix: passthrough --json/--jq/--template flags in all gh subcommands by shadowofdoom · Pull Request #196 · rtk-ai/rtk

shadowofdoom · 2026-02-18T13:15:25Z

Summary

gh pr view --json fields failed with Unknown JSON field: "--json" because view_pr() treated --json as the PR identifier
The v0.21.1 fix (fix: gh run view drops --log-failed, --log, --json flags #159) only covered gh run view via should_passthrough_run_view()
This generalizes the fix with has_output_format_flags() and applies it to all handlers that hardcode --json: list_prs, view_pr, pr_status, list_issues, view_issue, list_runs, and run_repo
Also fixes view_pr to handle omitted PR number (gh defaults to current branch)

Reproduction

rtk gh pr view --json number,title,state
# Before: Unknown JSON field: "--json"
# After: passes through to gh correctly

Test plan

All 383 tests pass (5 new tests for has_output_format_flags)
rtk gh pr view --json number,title,state passes through correctly
rtk gh pr list still uses RTK's compressed output
rtk gh pr view <number> still uses RTK's compressed output

🤖 Generated with Claude Code

view_pr() assumed args[0] was always a PR number, so `gh pr view --json fields` treated "--json" as the PR identifier and appended its own --json, causing `Unknown JSON field: "--json"`. The v0.21.1 fix (should_passthrough_run_view) only covered `gh run view`. This generalizes it with has_output_format_flags() and applies it to all handlers that hardcode --json fields: list_prs, view_pr, pr_status, list_issues, view_issue, list_runs, and run_repo. Also fixes view_pr to handle omitted PR number (gh defaults to current branch). Closes rtk-ai#159 (broader fix) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

pszymkowiak · 2026-02-20T15:30:44Z

Tested locally on the branch — LGTM, approve.

All 383 tests pass. Manual testing confirms:

gh pr view --json, gh pr list --json --jq, gh issue list --json --jq, gh repo view --json, gh pr status --json, gh run list --json --jq — all passthrough correctly with raw JSON output
gh pr list, gh pr view 196, gh issue view 228 — RTK filtering still works, no regression

Minor nits (non-blocking):

cargo fmt flags 2 formatting issues in tests
Consider dropping -t/-q short flags from has_output_format_flags() — they could cause false positives if other gh subcommands reuse those short flags for something else. --json, --jq,
--template are sufficient and unambiguous.

pszymkowiak · 2026-02-20T15:32:26Z

please review conflict to merge

…tk-ai#217, rtk-ai#196, rtk-ai#248, rtk-ai#211, rtk-ai#200, rtk-ai#192, rtk-ai#268) Wave 1 (critical bugs): - fix(registry): fi/done moved to IGNORED_EXACT — find no longer shadowed (rtk-ai#246) - fix(playwright): f64 duration, specs[] structure, --reporter=json after subcmd (rtk-ai#193) - fix(gh): should_passthrough_gh_view for --json/--jq/--template/--web in view_pr/issue/run (rtk-ai#217+196) Wave 2 (reliability): - fix(git): is_blob_show_arg — blob show passthrough without trailing-newline trim (rtk-ai#248) - fix(find): parse_find_args with native -name/-type/-maxdepth/-iname support (rtk-ai#211) - fix(main): graceful Clap fallback + parse_failures SQLite table + rtk gain --failures (rtk-ai#200) Wave 3 (UX): - feat(git): global options -C/-c/--git-dir/--work-tree/--no-pager/--no-optional-locks/--bare/--literal-pathspecs (rtk-ai#192) - feat(proxy): streaming output via spawn()+threads instead of buffered output() (rtk-ai#268) Tests: 1091 → 1117 (+26), 0 regressions Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

MickaelV0 · 2026-03-03T09:12:21Z

Ran into this exact issue while auditing RTK compatibility with a plugin-heavy Claude Code setup. Sharing some findings that might help justify prioritizing this merge.

What breaks in practice

We use a dev-core plugin (25+ skills) where Claude Code directly runs gh commands with structured output flags and then parses the result — either via inline --jq filters or by passing JSON to downstream programs. With RTK installed, these all silently break:

gh issue list --search "..." --json number,title,state --jq '.[:3]'
gh pr list --head "$BRANCH" --json number,state,reviewDecision --jq '.[]'
gh pr view --json comments --jq '.comments[].body'
gh repo view --json name --jq '.name'

The failure mode is subtle: the hook rewrites the command, RTK runs gh internally with its own hardcoded --json fields (prepended), then reformats the output as compact text. By the time --jq runs (as a gh flag, not a shell pipe), the fields requested by the user may not even be present. Then RTK reformats the final result anyway — Claude gets compact text where it expected a JSON array.

What we verified in the source

The hook pattern '^gh[[:space:]]+(pr|issue|run|api|release)' matches unconditionally on command name, with no flag inspection
list_issues() and view_issue() in gh_cmd.rs prepend their own --json fields before appending user args, then always reformat via serde_json
gh project subcommands correctly fall through to run_passthrough — that part works great

Workaround we're using in the meantime

From the thread in #188, @mariusvniekerk's stopgap (skip rewrite when piping to jq) handles shell-pipe cases. But it doesn't cover gh's built-in --jq flag or bare --json usage without a pipe. PR #196's has_output_format_flags() approach is the right fix.

Is there anything blocking this beyond the merge conflict? Happy to help test if a rebase lands.

MickaelV0 · 2026-03-03T09:23:18Z

Architectural position needed: structured-output flags should always passthrough

Beyond this specific fix, I think RTK needs a clear architectural stance on passthrough behavior for any CLI flag that signals structured/machine-readable output.

The issue I hit with `gh --json` / `--jq` is not a one-off — it's a fundamental pattern that will repeat for every proxied CLI that supports structured output conventions:

`gh` → `--json`, `--jq`, `--template`
`git` → `--porcelain`, `--format`, `--name-only`
`docker` → `--format`, `--quiet`
`kubectl` → `-o json`, `-o yaml`, `-o jsonpath`
etc.

When a Claude Code skill (or any plugin) runs a CLI command and parses the output programmatically — via inline `--jq`, downstream `jq` pipes, or direct JSON deserialization — RTK's reformatting silently breaks the contract.

Proposed convention: any CLI that has established object-passing or post-processing conventions (structured output flags, machine-readable modes) should be treated as passthrough when those flags are present, to guarantee correct integration with skill and plugin ecosystems.

PR #196's `has_output_format_flags()` is exactly the right pattern. Elevating it to a documented design principle — and applying it systematically to every future CLI proxy RTK adds — would prevent this entire class of breakage upfront, rather than fixing it case by case.

FlorianBruniaux · 2026-03-05T18:53:20Z

Hi @shadowofdoom! We have multiple gh flag passthrough PRs open (#196, #217, #319, #325, #328). They all conflict with each other. Could you rebase both PRs on latest master first to fix the CI failures, then we'll coordinate which approach to take? Alternatively, feel free to consolidate your two PRs into one — that would help.

gh commands with --json, --jq, or --template produce structured output that rtk gh filtering would corrupt. Skip rewrite so callers get raw JSON.

…LED, 2>&1, json TOML

pszymkowiak · 2026-03-05T22:29:14Z

Fixed in v0.27.0 — gh with --json/--jq/--template now skips rewrite.

gh commands with --json, --jq, or --template produce structured output that rtk gh filtering would corrupt. Skip rewrite so callers get raw JSON.

Conflict resolution (1 conflict, additive): - src/discover/registry.rs: keep all tests from both sides - ours: test_classify_wc, test_classify_wc_bare, test_route_wc - upstream: test_rewrite_gh_json_skipped, test_rewrite_gh_jq_skipped, test_rewrite_gh_template_skipped, test_rewrite_gh_api_json_skipped, test_rewrite_gh_without_json_still_works (rtk-ai#196) Upstream v0.27.0 changes absorbed: - fix(registry): RTK_DISABLED=1 env prefix skips rewrite entirely (rtk-ai#345) - fix(registry): gh --json/--jq/--template skips rewrite to avoid corrupting structured output (rtk-ai#196) - fix: RTK_DISABLED ignored, 2>&1 broken, json TOML error (rtk-ai#345,rtk-ai#346,rtk-ai#347) - docs: version refs, module count, CHANGELOG, ARCHITECTURE Also fixed pre-existing test race exposed by new parallel tests: - fix(test): add EnvGuard to test_shared_is_hook_disabled_* in claude.rs to prevent RTK_ACTIVE race with test_raii_guard_clears_on_panic (both tests set RTK_ACTIVE without holding ENV_LOCK) Tests: 1029 pass, 0 fail (parallel), 5 ignored

… smart, docker, json edge cases Covers bug fixes #196, #344, #345, #346, #347 and previously untested commands. Adds assert_fails helper. 118 assertions total (was 69).

…nged Bug: rtk hook claude was routing gh pr list --json ... to rtk gh pr list --json ... which corrupts structured JSON output. Mirrors upstream fix registry::rewrite_segment rtk-ai#196 to the binary hook path. Fix: should_passthrough() returns true for gh commands containing --json, --jq, or --template flags — hook emits no output (NoOpinion) so Claude Code runs the original gh command unchanged. Tests: 2 new (test_gh_json_flag_passes_through, test_gh_without_json_not_passthrough); 1031 pass total

…at/gemini-support-v2 Conflict resolution (1 conflict, additive): - src/main.rs Init command: merge --claude/--gemini flags (gemini branch) with --hook-type flag (rust-hooks-v2); keep all three flags - Dispatch: multi-platform logic calls init::run(..., hook_type, ...) for Claude and init::run_gemini() for Gemini; hook_type threads through Result: rtk init now supports --claude, --gemini, and --hook-type together rtk init → Claude + Gemini, script hook (default) rtk init --claude → Claude only rtk init --gemini → Gemini only rtk init --hook-type binary → Claude + Gemini with binary hook Inherits from rust-hooks-v2 (dbc46c7): - fix(hook): binary hook passes gh --json/--jq/--template through unchanged (mirrors upstream registry::rewrite_segment fix rtk-ai#196) - 2 new tests: test_gh_json_flag_passes_through, test_gh_without_json_not_passthrough Tests: 1050 pass, 0 fail (parallel), 5 ignored

* fix: prettier reports "All OK" when not installed (#221) Empty or failed prettier output was incorrectly treated as "all files formatted". Now detects empty output and non-zero exit code, shows the actual error message instead of a false positive. * test: add smoke tests for rewrite, verify, proxy, discover, diff, wc, smart, docker, json edge cases Covers bug fixes #196, #344, #345, #346, #347 and previously untested commands. Adds assert_fails helper. 118 assertions total (was 69). * chore: update benchmark.sh with missing commands and fix paths - Add cargo (build/test/clippy/check), diff, smart, wc, curl, wget sections - Fix Python commands: use dedicated rtk ruff/pytest instead of rtk test - Fix Go commands: use dedicated rtk go/golangci-lint, add go build/vet - Make BENCH_DIR absolute so debug files work from temp fixture dirs - Fallback to installed rtk if target/release/rtk not found

`rtk curl` rewrites every `curl URL` to `rtk curl URL` and pipes the response through `rtk json --schema`, which produces field-type literals (`field: int`) and a `(N)` array-length suffix. That's a token-savings win for human-facing exploration of arbitrary third-party APIs, but it ACTIVELY BREAKS downstream JSON parsing (jq, `python json.load`, agent-side filtering, anything that round-trips JSON) for private or internal APIs whose responses are consumed by parsers rather than read by humans. Today the only escape hatch is `[hooks] exclude_commands = ["curl"]`, which disables curl rewriting entirely — losing the token savings on the 3rd-party APIs that motivated the rewrite in the first place. This change adds a narrow opt-in mechanism: a substring allowlist that opts specific URLs out of the rewrite while leaving everything else rewriting as before. [curl] bypass_url_markers = [ "localhost:8080/api/", "//internal.example.com/v1/", ] Each marker is matched as a substring against the full command segment (after env-prefix stripping), so it composes cleanly with curl's flag positioning (`-X POST`, `-H`, `-d`, etc). Bypass is per-segment for compound commands: `a && curl <internal> | jq` bypasses the curl segment while `a` and the pipe target still rewrite as normal. Default `bypass_url_markers = []` preserves historical behavior — users who don't configure anything see no change. API surface ----------- * New `crate::core::config::CurlConfig { bypass_url_markers: Vec<String> }` * New `crate::discover::registry::RewriteOptions { curl_bypass_url_markers }` * New `rewrite_command_with_options(cmd, excluded, &opts)` — takes options explicitly, useful for tests and callers that already have config in hand. * Existing `rewrite_command(cmd, excluded)` is unchanged for callers but now reads `Config.curl.bypass_url_markers` internally and forwards. Mirrors the existing rtk-ai#196 bypass shape for `gh --json/--jq/--template`: detect a structured-output consumer and skip schema-mode filtering. Tests ----- * 4 new config tests (default empty / [curl] roundtrip / missing section). * 9 new registry tests covering: localhost / loopback / hostname marker bypass; POST with headers + payload; multi-marker OR; default-empty preserves rewrite; unmatched URL still rewrites; port-specific narrowness; per-segment behavior in compound commands. * Full suite: 1699 passed, 0 failed, 6 ignored.

gh commands with --json, --jq, or --template produce structured output that rtk gh filtering would corrupt. Skip rewrite so callers get raw JSON.

… --json, hook check, RTK_DISABLED, 2>&1, json TOML

) * fix: prettier reports "All OK" when not installed (rtk-ai#221) Empty or failed prettier output was incorrectly treated as "all files formatted". Now detects empty output and non-zero exit code, shows the actual error message instead of a false positive. * test: add smoke tests for rewrite, verify, proxy, discover, diff, wc, smart, docker, json edge cases Covers bug fixes rtk-ai#196, rtk-ai#344, rtk-ai#345, rtk-ai#346, rtk-ai#347 and previously untested commands. Adds assert_fails helper. 118 assertions total (was 69). * chore: update benchmark.sh with missing commands and fix paths - Add cargo (build/test/clippy/check), diff, smart, wc, curl, wget sections - Fix Python commands: use dedicated rtk ruff/pytest instead of rtk test - Fix Go commands: use dedicated rtk go/golangci-lint, add go build/vet - Make BENCH_DIR absolute so debug files work from temp fixture dirs - Fallback to installed rtk if target/release/rtk not found

- Add svn RtkRule to rules.rs so hook auto-rewrites svn status/log/diff/ info/blame/add/commit/update/up to rtk-prefixed equivalents - Add --xml detection to registry rewrite_segment (like gh --json in rtk-ai#196) to skip hook rewriting for machine-readable XML output - Add module-level --xml passthrough in svn run_log and run_info so that even direct `rtk svn log --xml` bypasses filtering Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>

…surable (rtk-ai#196) saved_tokens is computed with saturating_sub, so a filter that emits MORE tokens than it consumed records as "0% saved" and the regression vanishes from the headline stats. Telemetry showed 1,410 commands (11.5%) with output > input, +25,846 tokens net, all hidden as "0%". Add an inflation_tokens column (output - input, floored at 0) populated on every record(), plus a total_inflation_tokens() accessor. saved_tokens and savings_pct are deliberately UNCHANGED — making saved signed would break the unsigned SUM aggregations in gain.rs and the no_bloat leaked-tokens calc. Migration mirrors the existing exec_time_ms/project_path ALTER TABLE pattern (idempotent; duplicate-column error swallowed). Council design (Codex + mmax, prior round): separate column, do not touch saturating_sub, leave streaming as-is. Council review (this round): SHIP (Codex + mmax, no blocking findings). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…record) (rtk-ai#196) TimedExecution::track clamped output to `estimate_tokens(output).min(input_tokens)` before calling record(). record() derives inflation as `output_tokens.saturating_sub(input_tokens)`, so the clamp guaranteed inflation was ALWAYS 0 in production — the entire rtk-ai#196 column was dead on arrival. The unit test passed only because it called record() directly with un-clamped values, bypassing the clamp (a false green). Fix: drop the `.min(input_tokens)` clamp in track() and pass the real `estimate_tokens(output)`. record() already floors `saved` at 0 via `input.saturating_sub(output)`, so the no-bloat guarantee of #95 is preserved (inflating filter still reports 0 saved / 0% savings, never negative). inflation_tokens now captures the true overflow. track_passthrough passes 0,0 and is unaffected; it was the only other record() caller. Tests: replaced the clamp-asserting tests (which encoded the bug) with tests that assert the corrected behaviour, and added a regression driven through track() (not record() directly) proving an inflating filter records inflation_tokens > 0 while saved_tokens == 0 and savings_pct == 0 — the regression the old suite missed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

rtk-ai#196) The inflation_tokens column now records real data (the clamp that zeroed it was removed), but nothing user-visible exposed it. savings_pct floors at 0, so a filter that emits MORE tokens than the raw command reads as "0% saved" and the regression stays hidden. Surface it additively, with no change to clean-install output: - GainSummary gains total_inflation (SUM over inflation_tokens in the existing project-scoped summary query). gain prints a "Tokens inflated" KPI line ONLY when > 0, so clean installs are unaffected. - WeakFilter gains inflation_tokens (SUM(inflation_tokens) added to the existing GROUP BY rtk_cmd weak-filter query). gain --weak-filters adds an "Inflated" column; clean tools show a dash, inflating tools show the figure so a regression stands out. - JSON export carries total_inflation for machine-readable parity. Both queries reuse the established project-path filter / time-window patterns, so scoping stays consistent. Tests (in-memory tracker / RTK_DB_PATH pattern): summary aggregates inflation across rows; reports 0 on clean data (line omitted); weak filters carry per-tool inflation with a saving tool reporting zero. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

… follow-up) run_fallback recorded a parse_failure + a 0-saving `commands` row for every passthrough — including commands contextcrawler never owned (unknown subcommands like `cortextos bus`, `bash -lc ...`). That skewed the `gain` savings % (147 parse-failures + 173 zero-saving rows of pure cron traffic in real data) and flooded `discover` with non-actionable noise. Add is_foreign_command() = clap ErrorKind::InvalidSubcommand and, in the no-TOML passthrough branch, skip BOTH track_passthrough and record_parse_failure for foreign commands. They still pass through and propagate the real exit code — they're just not counted, because they were never a candidate for filtering. Genuine parse failures on commands we DO own (other error kinds) stay tracked; TOML-matched commands are unaffected. Reviewed by mmax (no blockers): InvalidSubcommand is the correct/sufficient signal, exit-code propagation unchanged, tracking is stats-only (no security impact). Test: is_foreign_command distinguishes unknown subcommand (foreign) from bad-args on a known command (real failure). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…(rtk rtk-ai#2035) The compact `git status` path ran `git status --porcelain -b -uall`, but we measure savings against raw `git status` — which defaults to `-unormal` and collapses fully-untracked directories to a single line. `-uall` expands them into every file, so in any repo with a non-ignored untracked dir our output exceeded raw (~29x on a 200-file dir), the opposite of the tool's purpose — and the inflation was invisible because `savings_pct` floors at 0 (rtk-ai#196). Drop `-uall` so untracked dirs collapse exactly like raw. Modified/staged/ renamed/conflict paths are unaffected by `-uall` and remain fully shown by the formatter; state-header and detached-HEAD extraction read from the separately-captured plain status, so they are unchanged. Aligns with upstream rtk PR rtk-ai#2035. Council: Codex + mmax both APPROVE. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Council (codex MEDIUM): the "vs raw output" line was shown only when total_saved != effective_saved. But a huge zero-savings dump — an rtk-ai#196 inflation row or an unfiltered passthrough — contributes 0 to BOTH saved totals while still shrinking effective_input, so the raw and effective percentages diverge (e.g. 0.007% vs 0.92%) yet the raw line stayed hidden. Gate on total_input != effective_input instead: that differs exactly when some command was capped, which is exactly when the two percentages can diverge. mmax SHIP; this closes codex's HOLD. Refs: rtk-ai#208 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01K3SAimSBwPUp4MkRXaYXSU

gh --json and gh api were passthrough-only (rtk-ai#196): converting JSON to a bare schema would destroy values, so rtk never touched them and tracked 0% savings on exactly the outputs where field names repeat per row. This adds core/jsonpack: a lossless re-encoding ported from the compaction stage of headroom-core (headroomlabs-ai/headroom, Apache-2.0) with every lossy path removed — no retrieval pointers, no row dropping, no stringified-JSON rewriting. - top-level arrays of objects render as a [N]{col:type} declaration plus CSV rows; empty cell = absent key, bare null = JSON null, lookalike strings stay quoted, mixed columns carry JSON literals - envelopes (gh api) stay valid JSON: dense inner arrays become {"_cols":…,"_rows":…} tables and the body is minified - pack() decodes its own output and requires value equality with the input before emitting; any mismatch (including data that collides with the notation) falls back to the raw bytes — Never Block - --jq/--template (long, short, = and attached forms) keep full passthrough; the registry rewrites bare --json to rtk gh again Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

gh --json and gh api were passthrough-only (rtk-ai#196): converting JSON to a bare schema would destroy values, so rtk never touched them and tracked 0% savings on exactly the outputs where field names repeat per row. This adds core/jsonpack: a lossless re-encoding ported from the compaction stage of headroom-core (headroomlabs-ai/headroom, Apache-2.0) with every lossy path removed — no retrieval pointers, no row dropping, no stringified-JSON rewriting. - top-level arrays of objects render as a [N]{col:type} declaration plus CSV rows; empty cell = absent key, bare null = JSON null, lookalike strings stay quoted, mixed columns carry JSON literals - envelopes (gh api) stay valid JSON: dense inner arrays become {"_cols":…,"_rows":…} tables and the body is minified - pack() decodes its own output and requires value equality with the input before emitting; any mismatch (including data that collides with the notation) falls back to the raw bytes — Never Block - --jq/--template (long, short, = and attached forms) keep full passthrough; the registry rewrites bare --json to rtk gh again Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: Rui Botte <nfsbotte@gmail.com>

`rtk rewrite 'gh pr view 123 --json reviews'` exited 1 with no output, so the hook passed the command through unrewritten — `gh pr view --json` is how gh is used in practice, so those calls were never tracked. The rtk-ai#196 guard assumed `rtk gh` would reshape structured output, but `gh_cmd::run` has passed `--json` straight to gh since then, the same stance `gh api` takes. Rewriting is therefore byte-for-byte safe and the guard only cost coverage. - drop the registry special-case for `rtk gh` + --json/--jq/--template - harden the passthrough gate it was masking: `has_json_flag` matched only a bare `--json`, so `gh pr view 42 --json=number` fell into the filter and returned rtk's summary instead of the requested field. `wants_raw_output` now covers --json/--jq/-q/--template/-t in both spellings - flip the rtk-ai#196 registry tests to assert the rewrite, add unit tests for the flag spellings Verified against a real PR: `gh pr view N --json number,title,state,reviews` and the --jq/--template forms are byte-identical through `rtk gh`. Fixes rtk-ai#4067 Signed-off-by: dopaemon <polarisdp@gmail.com>

pszymkowiak added the awaiting-changes label Feb 20, 2026

pszymkowiak mentioned this pull request Feb 20, 2026

feat(hook): native cross-platform (Windows & more) hook-rewrite command #150

Closed

6 tasks

This was referenced Mar 5, 2026

fix(gh): passthrough output when --jq flag is present #319

Closed

fix: passthrough gh commands when --json flag is present #325

Closed

fix: pass through -R/--repo flag in gh view commands #328

Merged

FlorianBruniaux mentioned this pull request Mar 5, 2026

fix: gh pr/issue/run view drops flags when no identifier given #217

Closed

5 tasks

pszymkowiak added a commit that referenced this pull request Mar 5, 2026

fix: skip rewrite for gh --json/--jq/--template (#196)

079ee9a

gh commands with --json, --jq, or --template produce structured output that rtk gh filtering would corrupt. Skip rewrite so callers get raw JSON.

pszymkowiak added a commit that referenced this pull request Mar 5, 2026

fix: bugs #196 #344 #345 #346 #347 — gh --json, hook check, RTK_DISAB…

8953af0

…LED, 2>&1, json TOML

github-actions Bot mentioned this pull request Mar 5, 2026

chore(master): release 0.27.0 #353

Merged

pszymkowiak closed this Mar 5, 2026

BrewTestBot mentioned this pull request Mar 6, 2026

rtk 0.27.0 Homebrew/homebrew-core#270865

Merged

pszymkowiak mentioned this pull request Mar 6, 2026

fix: prettier false positive when not installed (#221) #359

Merged

4 tasks

This was referenced Mar 11, 2026

warn: detect and report RTK_DISABLED=1 overuse in sessions #508

Closed

gh PR comments/reviews and git diff output still truncated, causing repeated fetch attempts by AI agents #354

Closed

officialasishkumar mentioned this pull request Apr 30, 2026

fix(rewrite): pass ls -O / -@ / -e through unchanged (#1627) #1640

Open

3 tasks

globalsecurepayments mentioned this pull request May 2, 2026

feat(curl): config-driven URL allowlist to bypass schema-mode rewrite #1677

Closed

thehoff pushed a commit to thehoff/contextcrawler that referenced this pull request May 14, 2026

fix: bugs rtk-ai#196 rtk-ai#344 rtk-ai#345 rtk-ai#346 rtk-ai#347 — gh…

95cc1a2

… --json, hook check, RTK_DISABLED, 2>&1, json TOML

rbotte mentioned this pull request Aug 15, 2026

feat(gh): lossless csv+schema packing for gh --json and gh api output #3580

Open

This was referenced Sep 14, 2026

fix(gh): rewrite gh --json/--jq/--template instead of deferring #4069

Open

rewrite: gh pr view --json <field> fails silently (exit 1, no rewrite) #4067

Open

Navigation Menu

fix: passthrough --json/--jq/--template flags in all gh subcommands - #196

fix: passthrough --json/--jq/--template flags in all gh subcommands#196
shadowofdoom wants to merge 1 commit into
rtk-ai:masterfrom
shadowofdoom:fix/gh-view-passthrough-json

shadowofdoom commented Feb 18, 2026

Uh oh!

pszymkowiak commented Feb 20, 2026

Uh oh!

pszymkowiak commented Feb 20, 2026

Uh oh!

MickaelV0 commented Mar 3, 2026

Uh oh!

MickaelV0 commented Mar 3, 2026

Uh oh!

FlorianBruniaux commented Mar 5, 2026

Uh oh!

pszymkowiak commented Mar 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

shadowofdoom commented Feb 18, 2026

Summary

Reproduction

Test plan

Uh oh!

pszymkowiak commented Feb 20, 2026

Uh oh!

pszymkowiak commented Feb 20, 2026

Uh oh!

MickaelV0 commented Mar 3, 2026

What breaks in practice

What we verified in the source

Workaround we're using in the meantime

Uh oh!

MickaelV0 commented Mar 3, 2026

Uh oh!

FlorianBruniaux commented Mar 5, 2026

Uh oh!

pszymkowiak commented Mar 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants