Open
Conversation
- install.sh now points to algolia/rtk at pinned v0.22.2 (overridable via RTK_VERSION env var) - All docs, scripts, CI workflows updated from rtk-ai/rtk to algolia/rtk - CHANGELOG historical upstream links preserved for attribution - All GitHub Actions workflows target 'main' branch (not 'master') - New ci.yml: build/test/clippy/fmt on ubuntu + macOS - New telemetry-guard CI job: blocks telemetry.rs, ureq, sha2, hostname deps, and phone-home patterns in source - CLAUDE.md documents fork maintenance strategy, sync policy, and telemetry exclusion rules
…re-commit hook to match CI
fork: establish algolia/rtk identity with no-telemetry CI guard
release: v0.22.3
feat: support git -C directory option
release: v0.22.4
gh: when --json flag is present, bypass all filtering and passthrough raw JSON output. RTK was reformatting structured JSON into lossy human-readable summaries, breaking downstream jq/python consumers. grep: detect -c/--count mode and passthrough rg output verbatim. The file:count format was being misinterpreted as file:linenum:content, producing gibberish like "📄 101 (1): prefix_saturated". Closes bug-report: 2026-03-19-gh-json-output-rewritten.md
scripts/fuzz-rtk.py uses Qwen 3.5 (via Algolia inference proxy) to generate diverse command invocations targeting format-changing flags, then compares raw vs RTK output with 6 heuristic checks: JSON integrity, line expansion, emoji injection, exit code match, data loss, and format preservation. First run: 64 tests, 30 failures (47% failure rate). Key findings: - grep: most rg flags crash with exit 2 (-l, --vimgrep, -A/-B/-C, --json) - git log: custom --format/--pretty output mangled by compact filter - git log: --graph characters stripped, --stat/--patch data lost Bug reports documented in bug-reports/2026-03-19-fuzz-findings.md
RTK's grep had short flags colliding with rg: -l (RTK: --max-len) vs rg -l (--files-with-matches) -c (RTK: --context-only) vs rg -c (--count) -m (RTK: --max) vs rg -m (--max-count) Removed colliding short flags from Clap. Now -l, -c, -m flow through to rg via extra_args as users expect. Generalized passthrough to all format-changing rg flags: -l, --json, --vimgrep, -h, --count-matches, and context flags (-A/-B/-C). These bypass RTK's grouping filter since they produce different output formats. Found via agentic fuzzing — 14/16 grep tests were failing before.
When user provides --format, --pretty, --graph, --stat, --numstat, --shortstat, --patch, -p, --raw, --name-only, or --name-status, RTK now passes output through verbatim instead of applying its compact filter. These flags indicate the user wants specific output formatting that RTK's compression would mangle. --oneline without other detail flags still gets RTK's default compression since it produces compatible output. Found via agentic fuzzing — 13/16 git-log tests were failing.
- Add 11 new command families: find, cat, tree, cargo-test, cargo-clippy, git-branch, git-stash, curl, wc, env, diff - Add STATIC_TESTS dict for deterministic regression without LLM - Extract _run_single_test() reusable helper - Switch to qwen3.5-35b-fp8 for faster throughput - Add FUZZ-RTK.md project README documenting architecture and usage
New families: git-diff, git-show, git-branch, find, cargo-build, cargo-clippy, diff, env, curl. Fix cargo 2>&1 safety false positive. Found 8+ new bugs across ls, find, diff, git-branch, git-show, wc.
…ind/ls/git-branch
Fixes discovered by agentic fuzzer (Run 2, bugs 8/10/11/12): - diff: exit code 1 when files differ (matches diff convention) - git branch: passthrough for --format/--sort flags - git show: passthrough for --name-only/--name-status/--raw/--no-patch/-p - ls: passthrough when -l flag present (preserves metadata users expect) Static fuzzer: 20 fail → 13 fail (65 tests)
- git diff: passthrough for --name-only/--name-status flags - cargo build/clippy/check: passthrough for --message-format=json - grep: disable_help_flag so -h flows to rg instead of Clap - fuzz-rtk.py: remove false cat -b test case
Maps -q to --brief flag, shells out to real diff for brief mode. Preserves exit codes (0=identical, 1=differ).
…rmat check - grep/rg: sort lines before FORMAT_ALTERED comparison to handle non-deterministic rg thread ordering (was causing false positives) - wc: skip FORMAT_ALTERED check entirely (filename stripping is intentional compression, not a format bug) Results: 47% → 8% failure rate (5 remaining are design choices)
…UZZ-005) Five bug classes discovered by LLM-powered fuzzing: - FUZZ-001: git format flags not passthrough - FUZZ-002: cargo --message-format=json mangled - FUZZ-003: git diff --name-only filtered incorrectly - FUZZ-004: grep -h intercepted by Clap - FUZZ-005: missing flag support (diff -q, ls -l, git branch)
New bug classes tested: - separator: -- handling between tool flags and args - large-output: truncation behavior on 100+ line results - stderr-commands: commands producing output on stderr - gh-issue/gh-api: GitHub CLI subcommands - empty-output: no-match/clean-state edge cases New heuristic (#7): STDERR_LOSS — detects when raw stderr content vanishes through RTK's filter pipeline. Preflight now skips vault/API checks for static-only runs. Multi-command families use RTK_MAP for command translation. Found 3 new bugs: cargo test -- separator loss, cargo clippy stderr loss, find -name Clap misinterpretation.
New families: docker-ps, docker-images, pip, go-test, npm, ruff, pytest, git-tag, git-remote, cargo-test (expanded), cargo-clippy (expanded). New bugs: FUZZ-006 (docker flags rejected), FUZZ-007 (pip format override), FUZZ-008 (npm hardcoded to run), FUZZ-009-012 (find, cargo stderr, git branch, diff data loss). LLM round on weak families: 94% failure rate confirms docker/pip/npm/find are comprehensively broken.
Add trailing_var_arg to DockerCommands::Ps and Images so Clap accepts extra flags. Format-changing flags (--format, -q, --quiet) trigger passthrough; content flags (-a, --filter, --no-trunc) pass through while RTK still filters the output. Also adds shared has_output_format_flag() utility to utils.rs for reuse across modules. Fixes 12 fuzzer failures.
…ZZ-007) When user specifies --format=freeze/columns/etc, skip injecting --format=json and passthrough raw output. Unknown subcommands now passthrough instead of erroring. Fixes format override for pip list/outdated. uv compatibility for edge flags (--not-required) is a separate uv issue.
…-008) Only npm run/run-script goes through the boilerplate filter. All other subcommands (list, outdated, config, view, install, etc.) passthrough to npm directly. Previously all input was prepended with "run", breaking npm list, npm config, npm --version, etc. Fixes 3 fuzzer failures.
Restructured Find command to use trailing_var_arg with manual parsing. When system-find predicates (-name, -type, -maxdepth, etc.) are present, delegates to system find. Without predicates, uses RTK's glob finder. Fixes 3 fuzzer failures where Clap rejected valid find predicates.
When cargo subcommand output was primarily on stderr (clippy, check), the filtered output was printed to stdout, causing STDERR_LOSS. Now detects the original channel and preserves it.
When user explicitly passes -a/--all or -r/--remotes, show full remote list instead of deduplicating against local branches. RTK's auto-added -a (when no list flag given) still deduplicates for token savings.
…9/gh) compact_diff() was called with max_lines=100 for PR diffs, silently dropping files beyond ~2 in multi-file PRs. Agents doing PR review received incomplete diffs with no truncation signal. Raised to 500. Includes regression test with synthetic 5-file PR diff.
Replace clap subcommand enum with flat trailing_var_arg for compose, then manually route subcommands after splitting global flags. Fixes: docker compose -f deploy/docker-compose.yml build
filter_curl_output now pretty-prints JSON with serde_json, keeping actual data values intact. Previous behaviour replaced values with type names (string, int) via json_cmd, destroying API response data.
Commands containing =$( or =` are now skipped entirely by the rewrite hook. The ENV_PREFIX regex misparsed VAR=$(grep ...) as an env prefix, mangling the command. Safer to skip than to risk broken shell syntax.
PLNech
force-pushed
the
main
branch
2 times, most recently
from
43f54e2 to
37ac0de
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Fixes
Remaining issues
Test plan