feat(glab): add GitLab CLI (glab) command support

[tnucera]

Summary

Adds token-optimized output for glab (GitLab CLI) commands, mirroring gh_cmd.rs patterns adapted for GitLab-specific differences. Integrated into the post-#826 module structure (src/cmds/git/).

Subcommands: mr (list/view/create/merge/approve/diff/note/update), issue (list/view), ci (list/status/trace), release (list/view), api (passthrough). Unhandled subcommands forwarded to native glab.

glab vs gh adaptations

Aspect	gh	glab
Notation	#42	!42
States	OPEN/MERGED/CLOSED	opened/merged/closed
Author	author.login	author.username
URL field	url	web_url
Body field	body	description
Merge check	mergeable	merge_status
CI status	statusCheckRollup	head_pipeline.status

Token savings

Command	Strategy	Savings
glab mr list	JSON -> compact text with state icons	up to 96%
glab mr view	JSON -> essential fields + filtered markdown	up to 93%
glab issue list	JSON -> compact text with state icons	up to 80%
glab issue view	JSON -> essential fields + filtered markdown	up to 80%
glab ci list	JSON -> status icons + ref	up to 82%
glab ci status	Text keyword parsing with raw fallback	up to 82%
glab ci trace	Strip ANSI, section markers, runner boilerplate	40-60%
glab release list	Tab parser (old glab) + raw fallback (glab 1.82+)	20-25%
glab release view	Strip SOURCES block, image lines, markdown noise	20-60%
glab api, glab *	Passthrough	tracked

Savings tested empirically (Apr 2026 smoke tests). Unit tests assert a floor of ≥60% on JSON filters and ≥20-30% on text filters.

Key design choices

• runner::run_filtered + RunOptions (mirrors gh_cmd.rs) via a local run_glab_json<F>() helper — unified exit-code propagation, tracking, and JSON-fallback-to-raw
• resolved_command("glab") everywhere (Windows .cmd wrapper compat, mirrors gh_cmd.rs)
• has_output_flag() guard at top of run() — passthrough when user passes -F/--output/--json
• should_passthrough_view() for --web, --comments on mr/issue view
• Unified mr_action() for merge/approve/note/update with extract_identifier_and_extra_args (handles glab mr note -m "msg" 42)
• -R/--repo and -g/--group declared at clap level, appended to args (not prepended — avoids breaking subcommand dispatch)
• Non-JSON fallback: when glab returns plain text (e.g. empty results), prints raw instead of failing
• ci status: text-based parsing (glab doesn't support -F json for this subcommand), with raw fallback for unrecognized output
• ci trace: strip ANSI codes, GitLab CI section markers, runner/git/artifact boilerplate; preserve build output and errors
• release view: strip SOURCES archive URLs, image lines, HTML comments, horizontal rules; collapse blank lines
• glab api: passthrough by design — API responses vary and filtering would destroy values
• Display: plain text tags ([open]/[merged]/[closed]), functional icons only, aligned with fix: remove decorative emojis from CLI output (#511) #558
• Markdown filtering copied into glab_cmd.rs (shared-module refactor deferred to a separate PR)

Files changed

• src/cmds/git/glab_cmd.rs (new) — all handlers + 54 unit tests (auto-discovered via automod::dir!)
• src/main.rs — Commands::Glab variant with -R/-g flags, routing, is_operational_command
• src/discover/rules.rs — glab rewrite pattern + RtkRule
• src/discover/registry.rs — "GitLab" => 200 in category_avg_tokens + 5 classification/rewrite tests
• tests/fixtures/ — glab_mr_list_raw.json, glab_issue_list_raw.json, glab_release_list_raw.txt, glab_release_view_raw.txt, glab_ci_trace_raw.txt

Closes #851, closes #1085. Partially addresses #122.

Test plan

• cargo fmt --all --check — clean
• cargo clippy --all-targets — 0 new warnings
• cargo test --all — 1665 passed, 6 ignored
• Manual smoke tests against a live GitLab instance (2 projects, with -R cross-project targeting)
• Tested: mr list/view/diff, issue list, ci list/status, release list/view, api passthrough, ultra-compact mode, -F json passthrough guard, empty results fallback, 404 error propagation

[tnucera]

Manual smoke test results

Tested against a live GitLab instance with multiple projects.

Without -R (local repo context)

Command	Result	Notes
rtk glab mr list	✅	Compact table with state icons
rtk glab mr view <N>	✅	JSON extraction + markdown filtering
rtk glab mr diff <N>	✅	Diff compaction via git::compact_diff
rtk glab ci status	✅	Current branch pipeline status
rtk glab ci list	✅	10 pipelines with status icons
rtk glab ci view	✅	Passthrough (interactive TUI, cannot capture)
rtk glab pipeline list	✅	Alias for ci list, works correctly
rtk glab issue list	✅	Graceful fallback when no issues (plain text)
rtk glab api /projects/<id>	✅	JSON schema filtering
rtk glab repo view	✅	Passthrough for unknown subcommands

With -R owner/repo (cross-project)

Command	Result	Notes
rtk glab -R owner/repo mr list	✅	Correct project targeting
rtk glab -R owner/repo mr view <N>	✅	Correct MR from target project
rtk glab -R owner/repo mr diff <N>	✅	Diff from target project
rtk glab -R owner/repo ci list	✅	Pipelines from target project
rtk glab -R owner/repo ci status	✅	Correct error when no pipeline on branch
rtk glab -R owner/repo issue list	✅	Issues from target project
rtk glab -R owner/repo issue view <N>	✅	Full issue details from target project

With -g group (cross-group)

Command	Result	Notes
rtk glab -g group mr list	✅	MRs from all projects in group (20+, cap applied)
rtk glab -g group issue list	✅	Issues from all projects in group (plain text fallback)

Action commands (tested on a dedicated test project)

Command	Result	Notes
rtk glab mr create --source-branch ... --title ...	✅	ok created !2 <url> — MR created (verified with mr view)
rtk glab mr note <N> -m "..."	✅	ok noted !1 — comment added
rtk glab mr update <N> --title "..."	✅	ok updated !1 — title changed (verified with mr view)
rtk glab mr approve <N>	✅	401 self-approve restriction propagated correctly (GitLab limitation)
rtk glab mr merge <N> --yes	✅	ok merged !1 — MR merged (verified with mr view → state: merged)

Edge cases — empty project (0 MRs, 0 issues, 0 pipelines)

Command	Result	Notes
rtk glab -R owner/repo mr list	✅	Header only, no crash
rtk glab -R owner/repo issue list	✅	Graceful plain text fallback
rtk glab -R owner/repo ci list	✅	Header only, no crash
rtk glab -R owner/repo mr view 1	✅	glab 404 error propagated, exit code 1

Hook rewrite tests (rtk-rewrite.sh)

Input command	Rewritten to	Result
glab mr list	rtk glab mr list	✅
glab ci status	rtk glab ci status	✅
glab issue view 42	rtk glab issue view 42	✅
glab pipeline list	rtk glab pipeline list	✅
glab api /projects	rtk glab api /projects	✅
glab auth login	(not rewritten)	✅
glab repo clone foo	(not rewritten)	✅

Token savings observed

Command	Savings
glab mr list	49-99% (varies with result count)
glab mr view	77-94%
glab mr diff	36-98% (varies with diff size)
glab ci list	76-95%
glab api	~86%

Bugs found and fixed during testing

1. Non-JSON fallback: glab returns plain text (not JSON) when there are no results (e.g. empty issue list). Added graceful fallback in all JSON-parsing handlers.
2. -R flag not parsed by clap: -R/--repo and -g/--group were not declared in the Commands::Glab clap definition, causing the command to fall through to run_fallback. Fixed by adding them as explicit #[arg] fields (same pattern as git -C).
3. mr_view/issue_view not forwarding extra args: Functions that extract a number from args[0] were not passing remaining args (including -R) to glab. Added the missing for arg in &args[1..] loop.
4. ci view TUI crash: glab ci view launches an interactive TUI (tcell/tview) that panics when stdout is captured. Moved to passthrough (inherited stdio).
5. mr_action wrong subcommand: mr note and mr update passed the confirmation label ("noted", "updated") as the glab subcommand instead of the actual command ("note", "update"). Split into separate subcmd and label parameters.

[FlorianBruniaux]

Hi @tnucera, good structural work. Hook rule, registry entry, and CHANGELOG are all present, making this the most complete of the CLI integration PRs on that front. Two things to address:

1. discover/registry.rs declares savings_pct: 82.0 and subcmd savings for mr (87%), issue (80%), but there are no count_tokens() assertions in glab_cmd.rs verifying these numbers. Please add tests with JSON fixture inputs for at least mr_list and issue_list.

2. Throughout glab_cmd.rs, output is written to stdout piece-by-piece inside the formatting loop (print!("{}", line)) while also being accumulated into filtered for tracking. Building the full string first and printing once at the end (the pattern used in gh_cmd.rs) is cleaner and avoids the two representations diverging.

Missing tee integration is also worth noting, though it's not a blocker for a first merge.

[pszymkowiak]

Note: this PR adds rules to hooks/rtk-rewrite.sh, but since #241 (rtk rewrite), all rewrite logic lives in Rust (src/discover/rules.rs + src/discover/registry.rs). The bash hook now just calls rtk rewrite "$cmd".

Please remove the changes to rtk-rewrite.sh and add your glab patterns to rules.rs / registry.rs instead.

[tnucera]

Review feedback addressed

Rebased on upstream/master (includes PR #241 rtk rewrite) and addressed both reviews:

@FlorianBruniaux

1. count_tokens() tests with JSON fixtures — Added 7 new tests in glab_cmd.rs:

   • test_mr_list_token_savings / test_issue_list_token_savings — verify ≥60% savings with realistic 10-item JSON fixtures
   • test_mr_list_format / test_issue_list_format — verify output contains expected icons and IDs
   • test_mr_list_ultra_compact — verify compact mode
   • test_format_mr_list_invalid_json / test_format_issue_list_invalid_json — edge cases
   • Fixtures: tests/fixtures/glab_mr_list_raw.json and tests/fixtures/glab_issue_list_raw.json

2. Refactored output building — All handlers (mr_list, mr_view, issue_list, issue_view, ci_list, ci_status) now build the full filtered string first, then print!("{}", filtered) once at the end. Extracted format_mr_list_from_json() and format_issue_list_from_json() as pure functions (no I/O) for testability.

@pszymkowiak

1. Removed hooks/rtk-rewrite.sh changes — Accepted upstream version (thin delegator from PR feat: rtk rewrite — single source of truth for LLM hook rewrites #241).
2. Added glab patterns to rules.rs — Pattern r"^glab\s+(mr|issue|ci|pipeline|api)" + RtkRule with rewrite_prefixes: &["glab"], savings 82% (mr: 87%, ci: 82%, issue: 80%).
3. Updated registry.rs — Added "GitLab" => 200 in category_avg_tokens + 6 tests (test_classify_glab_mr, test_classify_glab_ci, test_classify_glab_issue, test_classify_glab_pipeline, test_rewrite_glab_mr_list, test_rewrite_glab_ci_status).

Quality gates

cargo fmt --all --check  ✅
cargo clippy --all-targets  ✅ (0 new warnings)
cargo test --all  ✅ 739 passed, 2 ignored

Live smoke tests (private GitLab instance)

Command	Result
rtk glab mr list	✅ Compact MR table with state icons
rtk glab mr view <N>	✅ JSON extraction + markdown filtering
rtk glab mr diff <N>	✅ Diff compaction
rtk glab issue list	✅ Graceful fallback (0 issues)
rtk glab ci list	✅ Header only (0 pipelines)
rtk rewrite "glab mr list" → rtk glab mr list	✅ rules.rs integration works

[tnucera]

Polish pass on glab_cmd.rs (post-audit, same commit)

Two fixes to align with gh_cmd.rs patterns:

1. run_api → passthrough: API responses are now passed through unchanged (like gh_cmd.rs:1204-1209). Converting JSON to a schema destroyed values and forced Claude to re-fetch — defeating the purpose. Removed json_cmd import.

2. ci_status fallback: Replaced raw.clone() with early return printing &raw directly — avoids a wasteful allocation on unrecognized formats.

Net: -33 lines, 739 tests passing.

[tnucera]

Hey! Quick update on this PR:

1. Rebased on latest master (0.30.1) — no conflicts
2. Aligned with PR fix: remove decorative emojis from CLI output (#511) #558 conventions — all decorative emojis removed ([open]/[merged]/[closed] instead of 🟢/🟣/🔴, plain text headers, etc.). Functional icons (✅/❌) preserved.
3. PR description updated to reflect the current state — please re-read it for the full picture

All quality gates pass locally: cargo fmt clean, cargo clippy 0 new warnings, cargo test 986 passed. Tested manually against a live GitLab instance.

@FlorianBruniaux @pszymkowiak Is there anything else needed to get this merged? Happy to address any feedback.

[jbdelhommeau]

We need it ! Thank you @tnucera

[CLAassistant]

All committers have signed the CLA.

[pszymkowiak]

Hi! Two things needed before we can review:

1. Retarget to develop — this PR targets master, but all PRs should target develop. You can change the base branch in the PR settings (right sidebar).
2. Sign the CLA — if not already done, please sign at https://cla-assistant.io/rtk-ai/rtk

Thanks!

[aeppling]

Hey

We are cleaning up the codebase and improving the project structure for better onboarding. As part of this effort, PR #826 reorganizes src/ from a flat layout into subfolders.

No logic changes — only file moves and import path updates.

What you need to do

Rebase your branch on develop when receiving this comment:

git fetch origin && git rebase origin/develop

Git detects renames automatically. If you get import conflicts, update the paths:

use crate::git;        // now: use crate::cmds::git::git;
use crate::tracking;   // now: use crate::core::tracking;
use crate::config;     // now: use crate::core::config;
use crate::init;       // now: use crate::hooks::init;
use crate::gain;       // now: use crate::analytics::gain;

Need help rebasing? Tag @aeppling

[tnucera]

Rebased on develop (post-refactoring #826) and addressed all review feedback:

• Retargeted to develop (@pszymkowiak)
• Removed rtk-rewrite.sh changes — glab patterns added to rules.rs / registry.rs instead (@pszymkowiak)
• count_tokens() tests added with JSON fixtures for mr_list and issue_list (>=60% savings verified) (@FlorianBruniaux)
• Output build refactored — all handlers build full string first, print once (@FlorianBruniaux)
• Adapted to feat(refacto-codebase-onboarding): partie 1 - folders and technical docs #826 module structure — glab_cmd.rs lives in src/cmds/git/ (@aeppling)
• Code review fixes: resolved_command() for Windows compat, has_output_flag() guard, should_passthrough_view() for --web/--comments, unified mr_action with proper identifier extraction, consistent from_str JSON parsing
• CLA signed

Quality gates: cargo fmt clean, cargo clippy 0 new warnings, cargo test 1161 passed.
Tested against a live GitLab instance with -R cross-project targeting.

[tnucera]

Hi @pszymkowiak, @FlorianBruniaux, @aeppling 👋

Just rebased this PR on latest develop — it was conflicting after the big refactors that landed recently (automod::dir!, Result<()> → Result<i32> exit-code contract, PATTERNS/RULES merge in discover/rules.rs). All three conflicts are resolved and glab_cmd.rs has been migrated to the new exit-code contract.

Current status:

• ✅ cargo fmt --all --check clean
• ✅ cargo clippy --all-targets — 0 new warnings
• ✅ cargo test --all — 1657 passed, 6 ignored, 0 failed
• ✅ End-to-end smoke tests against a live GitLab instance covering mr create → view → note → update → diff → merge — all handlers working, 76–96% token savings on read operations

What's missing to get this merged? 🙏

[tnucera]

Quick follow-up to tighten things before review:

• Refactored glab_cmd.rs to use runner::run_filtered + RunOptions + a local run_glab_json<F> helper, matching gh_cmd.rs architecture (net -221 LOC, unified tracking and exit-code flow)
• Added 8 edge-case tests (null nested fields, empty JSON arrays for mr/issue/ci, missing description, non-English CI trace fallback, no-SOURCES release view) — 54 unit tests total
• Fixed a latent bug in ci status fallback: now correctly preserves raw output when no English status keyword is recognized (previously the raw was silently replaced by a " "-prefixed version)
• Updated PR description: corrected test counts (1665 passed, 54 unit tests) and file list post-rebase

cargo fmt --all && cargo clippy --all-targets && cargo test --all all clean. Smoke tested end-to-end against a live GitLab instance (full mr create → view → note → update → diff → merge cycle). Ready for review 🙏

[aeppling]

Hey @tnucera , thanks for following our guidelines and codebase architecture updates.

Here is a review on filter quality, to ensure no important signal lost by aggressive filtering.
This is a main concern for us because loss of signal = retry = more token sent

Could you please verify those to ensure we do not integrate invalid filter ?

Filters quality

mr_view drops source/target branch info

The format_mr_view output has: state, title, author, merge_status, pipeline, URL, description. But it omits source_branch/target_branch. An LLM working on a merge workflow needs to know which branch is being merged into which. The gh_cmd.rs format_pr_view similarly doesn't show branches, but GitHub's PR model makes this less critical (the URL encodes it). For GitLab MRs, the branch info is often the most actionable signal.

mr_view drops labels and reviewers

Labels (e.g., bug, enhancement, blocked) and reviewer assignments are decision-relevant for an LLM asked to review/merge/triage MRs. The format_mr_list also drops labels — less critical there since it's a summary, but mr view is the detailed view.

Suggestions

pipeline_icon uses emoji for non-ultra-compact mode

"✅" and "❌" are multi-byte characters. The gh_cmd.rs may uses text tags [pass]/[fail] for non-compact mode.

Tech docs may be interesting for others contributors

This is a new integration and some technical documentation in the cmds/git/README.md could be interesting for futur contributors.
If you could check against others components of cmds/git/ if anything should be documented to explain any important architecture specifications, this could help futur contributors on this !

Thanks

For contributing to RKT and adding more coverage !

Others maintainers already reviewed, awaiting for their approvals since i've only reviewed on the surface :)
Should be released soon