---

name: running-tend description: Worktrunk-specific guidance for tend CI workflows. Adds codecov polling, Rust test commands, labels, and review criteria on top of the generic tend-* skills. Use when operating in CI. metadata: internal: true

Worktrunk Tend CI

Project-specific guidance for tend workflows running on worktrunk (a Rust CLI for managing git worktrees). The generic skills (tend-running-in-ci, tend-review, tend-triage, etc.) provide the workflow framework; this skill adds worktrunk conventions.

Filing issues in other repos

Standing exception granted: file directly in agent-equipped targets (per Filing Issues in Other Repos in the bundled running-in-ci skill) without asking permission here first. The default rule (open an issue here asking permission first) still applies when the target shows no agent signals.

Codecov Monitoring

After required CI checks pass, poll codecov/patch — it is mandatory despite being marked non-required:

for i in $(seq 1 5); do
  CODECOV=$(gh pr checks <number> 2>&1 | grep 'codecov/patch' || true)
  if echo "$CODECOV" | grep -q 'pass'; then
    echo "codecov/patch passed"; exit 0
  elif echo "$CODECOV" | grep -q 'fail'; then
    echo "codecov/patch FAILED"; exit 1
  fi
  sleep 60
done

If codecov fails locally, investigate with task coverage and cargo llvm-cov report --show-missing-lines | grep <file>.

Investigating codecov failures in CI

task and cargo-llvm-cov are not installed in the tend-setup action. Don't try to cargo install them in the sandbox — past attempts at source-compiling installs cascaded into bash-tool interrupts that blocked even pwd and echo. Instead, query Codecov directly, following tests/CLAUDE.md → Coverage Investigation for the endpoints and their traps. The sandbox also mounts root /tmp read-only, which is why the scratch paths there and below go to ${TMPDIR:-/tmp} — write new ones the same way.

If the Codecov API markers aren't enough, download the code-coverage-report artifact from the PR head's coverage workflow run — it contains a cobertura.xml with per-line hit counts:

REPO=$(gh repo view --json nameWithOwner --jq '.nameWithOwner')
# Find the coverage run on the PR head SHA:
CI_RUN=$(gh api "repos/$REPO/commits/<sha>/check-runs" --jq '.check_runs[] | select(.name == "code-coverage") | .details_url | capture("runs/(?<id>[0-9]+)") | .id')
# List artifacts, then download the coverage one:
gh api "repos/$REPO/actions/runs/$CI_RUN/artifacts" --jq '.artifacts[] | {name, id}'
gh api "repos/$REPO/actions/artifacts/<id>/zip" > "${TMPDIR:-/tmp}/coverage.zip"
unzip -q "${TMPDIR:-/tmp}/coverage.zip" -d "${TMPDIR:-/tmp}/coverage"

Test Commands

cargo run -- hook pre-merge --yes   # full suite + lints
cargo test --lib --bins             # unit tests only
cargo test --test integration       # integration tests only

CI runs on Linux, Windows, and macOS.

Session Log Paths

The artifact directory is named after the agent's working directory, and from tend 0.2.5 that is a per-run /tmp/tend-agent-workspace-*/checkout — so the old -home-runner-work-worktrunk-worktrunk/ prefix appears only in runs predating the bump, and there is no literal to match on any more. Use the bundled find "$DEST" -name '*.jsonl' recipe; either shape is one <session-id>.jsonl under a single slugified directory.

Labels

• automated-fix — fix PRs from triage and ci-fix workflows
• nightly-cleanup — nightly sweep issues and PRs

CI Fix: Prefer Rerun for Transient Infrastructure Failures

Before opening a fix/ci-* PR, classify the failure:

• Transient infrastructure (link-check timeouts, apt-get flakes, GitHub outages, runner disk issues, codecov upload blips) — do not create a PR. The maintainer will rerun CI. Comment on the run or exit silently; a permanent config change for a one-off timeout is churn the maintainer will close.
• Flaky test (known-flaky or first-seen PTY/shell test) — try to fix it.
• Real regression — proceed with a fix PR.

Non-required ≠ transient. A non-required job (e.g. collect affected coverage, affected tests (linux, advisory)) can fail from a real regression. The required/non-required distinction is about merge-blocking, not about how the failure is classified. If a deterministic build error (error[E...], "binary not found", "ambiguous candidates", missing target) repeats across consecutive runs of the same shape, it's a real regression even when the job is advisory. Reserve "transient" for non-deterministic causes: BrokenPipe, connection reset, runner disk full, GitHub API timeouts, host-availability blips.

Lychee link-check timeouts are always transient unless the same URL has failed on at least two separate runs within the last few days. The check runs as the link-check job in the nightly workflow. .config/lychee.toml already sets max_retries = 6 and lists known-unreliable hosts; one timeout is not enough evidence to extend that list. Signals you have a transient failure, not a broken link:

• The previous run on the same or a nearby commit passed.
• Only [TIMEOUT] is reported (not 404/403/410).
• The URL is reachable from a local curl.

When in doubt, post a comment on the failed run summarizing the diagnosis and wait — don't open a PR.

Applying GitHub Suggestions

Apply the literal suggestion only — change the lines it covers, nothing more. If surrounding lines also need updating, note that in your reply.

PR Review: Don't Self-Dismiss Over Unrelated Test Flakes

If a clearly-unrelated test fails after you've already approved a PR, leave the approval in place and post a comment noting the flake. Do not dismiss your own approval to "gate" on a rerun.

GitHub blocks both gh run rerun --failed and per-job rerun (POST /repos/{owner}/{repo}/actions/jobs/{id}/rerun) with HTTP 403 while any job in the same workflow run is still in_progress. The non-required benchmarks job routinely runs 80+ minutes after test (linux|macos|windows) finish, so dismiss-then-wait-then-rerun cascades into a long session for no benefit — the maintainer can rerun the failed job directly once benchmarks clears, or merge regardless if the failure is clearly a flake.

The codecov-failure dismissal pattern is different and remains correct: CLAUDE.md requires explicit user approval before merging with failing codecov/patch, so dismissing the approval until the coverage gap is addressed is intentional.

Weigh the root-cause fix before shipping a workaround

When a mismatch, a false positive, or a stale value has an obvious non-code workaround (a template change, a config value, an alias, a comment recording the drift), don't stop there. First check whether the workaround is lossy or foot-gunny, and weigh a proportionate root-cause code fix before opening a PR that only records it. A "docs-only, no risk" framing is not the same as good guidance — a zero-code-risk change can still steer users toward a collision-prone or lossy config, and annotating a stale value leaves the duplication that made it stale. If you do recommend a workaround, surface its downsides in the PR body up front, not only when challenged.

This governs every workflow that opens a PR here, not just issue triage.

Issue Triage

When you need more information to diagnose a reported bug, the primary ask is wt -vv <command>. Re-running the failing command with -vv writes a diagnostic bundle — a single report containing wt/git/OS versions, shell integration, wt config show, git worktree list --porcelain, and a trace.log of every git invocation with its output. The -vv output prints the bundle's exact absolute path (Diagnostics and performance profile saved @ …) followed by a ready-to-run gh gist create --web <path> line — point the user at those printed lines; don't hand them a hardcoded path. In particular, never tell them to cat .git/wt/logs/diagnostic.md: inside a linked worktree .git is a gitdir file, not a directory, so that path fails with Not a directory (os error 20) (ENOTDIR). The bundle actually lives under the git common dir — "$(git rev-parse --git-common-dir)/wt/logs/diagnostic.md" resolves from any worktree, and the printed absolute path already points there. One gist URL pasted into the issue gives us most of what we'd otherwise ask for piecemeal, so lead with this for unexplained failures rather than chaining version/config/repro questions across multiple round-trips.

When the report is about a slow wt command, read its Performance profile section first. It renders the same breakdown as wt config state logs profile (subprocess time by command type, slowest calls, repeated (command, context) pairs) directly from the bundled trace.log, so you can spot redundant git calls and slow commands without parsing the raw trace by hand. The same report run against a statusline capture is the weekly per-render cache check — see Weekly Maintenance: Statusline Cache-Check.

Reach for narrower asks only when the diagnostic is overkill:

• wt --version — when the only question is whether a fix has landed.
• wt config show — when the suspicion is purely config/shell-integration and you already have the command + repro.

Don't ship fixes you can't verify

When the bug or proposed fix turns on runtime state the bot can't observe from CI — plugin hooks firing inside an agent CLI (Claude Code, Codex, Gemini), shell-integration side effects, interactive prompt rendering, signal forwarding into a TTY — do not open a PR premised on the hypothesis. Signals to stop:

• The proposed transition fires inside a running agent session the bot can't drive from a test (PostToolUse, Stop, Notification, statusline redraws).
• The "analysis" in the issue is an LLM-written trace pasted by the reporter, not a verified observation. Treat that as a starting hypothesis, not ground truth — a Claude-written explanation of why X is broken is no more trustworthy than the bot's own first guess.
• The repro requires an interactive shell or claude running in a tmux that the bot can't spin up.

Comment on the issue with what's known, ask the reporter for the concrete symptom they observe ("which marker shows where, when") rather than for a fix to validate, and exit without a PR. The bar for opening a fix PR is the failure mode is reproducible and the fix is testable, not the hypothesis seems plausible. If you post a fix despite limited testability (rare — usually only when the reporter has confirmed the exact symptom and the code change is obviously correct from inspection), explicitly flag what wasn't verified in the PR body.

Closing Duplicates

When an issue is clearly a duplicate, close it after commenting. Use gh issue close <number> and tell the reporter: if they believe this was closed in error, they can let us know and we'll reopen it.

Check --config-set before calling a setting unpinnable

wt --config-set '<toml>' overrides any user config key for one invocation, above both config files and WORKTRUNK_SECTION__KEY env vars (inline config overrides). The resolver just reads repo.config().<key> and shows no sign of that layer, so before writing "there's no way to do X per invocation", ask whether X is a config key: the ask is usually already met, and the reply is a --config-set recipe plus a docs fix where the docs don't connect the flag to the setting. Pinning is all that covers — weigh the request's residual asks on their own merits.

Suggesting Aliases for Niche Feature Requests

worktrunk deliberately limits flag and config growth, so a wt alias is the standing answer to a narrow feature request rather than a new native flag. Suggest one when the request serves a single reporter's workflow or a small subset of users (idempotent create-or-switch, auto-push after merge) and composes from existing wt commands.

Answer with a wt [aliases] entry. Defined in user config, it resolves wt <name> to the alias whenever no built-in matches, so the same alias works across every repo. It's the project's preferred extension point.

How to respond:

1. Search open and closed issues for the same request and link the prior thread; these asks recur, and the link lets the deflection read as a considered position rather than a brush-off.
2. Draft the alias.
3. Test it in a scratch worktree against the happy path and edge cases (branch already exists, dirty worktree). When a surprising behavior turns up, note it in the reply instead of building the alias around it. For example, a create-or-switch wrapper inherits wt switch <name>'s habit of materializing a remote branch of the same name.
4. Post the tested alias with usage examples.
5. Link to the aliases docs and tips & patterns.

Don't fix tests by adding skip guards

When a test fails because production code or test setup can't handle some scenario, fix the production code or rework the test setup. Don't add an early-return skip — that removes the safety net while looking like a fix. If a triage fix reaches for let Ok(_) = ... else { return };, a newly-added if !path.exists() { return; }, or a fresh #[ignore], stop and ask what production behavior is actually broken.

If the test relies on inherited environment (process CWD, ambient env vars), rework it to set up its own — most worktrunk tests already do this via TestRepo::with_initial_commit() plus a tempdir.

Same-root-cause-class triage

The "work on the existing PR if it addresses the same problem" rule keys on the same test. It doesn't catch a different test failing for the same underlying reason. Group failing tests by root-cause class before writing a fix; if an outstanding PR addresses any test in the class, wait for it to merge and re-run, then mirror its approach for any sites still failing rather than opening a parallel PR with a weaker fix.

Weekly Maintenance: MSRV & Toolchain

Bump both MSRV and the development toolchain to latest stable − 1. When Rust 1.N is the current stable release, set both to 1.(N−1).

Files to update:

| File | Field | Example (if stable is 1.94) | |------|-------|----| | Cargo.toml | rust-version | "1.93" | | tests/helpers/wt-perf/Cargo.toml | rust-version | "1.93" | | rust-toolchain.toml | channel | "1.93.0" | | .github/workflows/nightly.yaml | rustup override set nightly-<date>, twice (minimal-versions, check-unused-dependencies) | a nightly from the last few weeks |

Bump the nightly pins only when the pinned date is more than three months old. Cargo refuses a workspace whose rust-version exceeds the toolchain, so an MSRV bump past a stale pin fails both jobs before they check anything.

flake.nix reads the channel from rust-toolchain.toml, so no separate bump is needed. After updating the toolchain, refresh flake.lock so the locked rust-overlay revision knows about the new version. tend-setup installs Nix with flakes enabled:

# Name the input: a bare `nix flake update` also relocks nixpkgs, an
# unrelated bump in a toolchain-scoped PR.
nix flake update rust-overlay
# Check the bumped channel still evaluates
nix eval .#devShells.x86_64-linux.default.name

Commit flake.lock alongside the other toolchain changes once both commands succeed. A failure is reported in the PR, never worked around: leave the file alone if the update fails, git checkout flake.lock if the eval does, and hand-compute an entry in neither case.

Expect both commands to fail with a permission error from tend 0.2.5 on. nix resolves on the agent's PATH, but the multi-user client reaches the store by connecting to /nix/var/nix/daemon-socket/socket, and the sandbox blocks socket(AF_UNIX, …) outright. That is the reported failure above, not a problem with the bump: say so in the PR, leave flake.lock untouched, and carry the rest of the toolchain change as normal. It clears when max-sixty/tend#1197 gives the action a lever for it.

After bumping, run the full test suite (cargo run -- hook pre-merge --yes) and verify cargo msrv verify passes.

Weekly Maintenance: CI Pin Bumps

Pinned third-party versions in CI are invisible to Dependabot — it follows Cargo.toml deps and uses: foo@vN action refs, not inline version: strings. They drift unless this step bumps them.

Each weekly run checks every pin below against upstream and bumps whatever has drifted. A bump that can turn a whole CI leg red — an OS image, a major version, a tool whose version moves snapshots — goes on its own branch and PR, so a red matrix decides only that bump instead of holding back the week's safe ones. The weekly runner is Linux, so that PR's own checks are the only place a macOS or Windows change gets tested; open it and report what they said.

• baptiste0928/cargo-install@v3 blocks in .github/workflows/{affected,ci,coverage,nightly}.yaml and .github/actions/{test,tend}-setup/action.yaml — every version: "=X.Y.Z" against cargo info <crate>. Today: cargo-affected, cargo-insta, cargo-nextest, cargo-llvm-cov, cargo-msrv, cargo-udeps, lychee, worktrunk. cargo-affected is pinned twice in affected.yaml; move both together. Verify each crate's rust-version against the pinned toolchain and note compatibility in the PR body (see PR #1657 for the format).

• hustcer/setup-nu@v3 version: input — latest from gh api repos/nushell/nushell/releases/latest --jq '.tag_name'. Five call sites: coverage.yaml (code-coverage), nightly.yaml (feature-powerset), benchmarks.yaml (benchmarks), and .github/actions/{test,tend}-setup/action.yaml — tend-setup's copy is what puts nu in the agent's sandbox, so it moves with the others.

• Codex Cloud tools — .codex/cloud.sh pins pre-commit, cargo-insta, cargo-nextest, Nushell, and PowerShell; setup-web pins Nushell and PowerShell. Keep cargo-insta, cargo-nextest, and Nushell level with .github/actions/test-setup/action.yaml, which pins the same three — the gate runs --all-features, so Nushell's version moves PTY snapshots. Nothing under .github/ pins PowerShell (CI runs whatever the runner image ships), so bump that one on its own.

• Docs site packages in docs/package.json are covered by Dependabot's /docs npm entry. Do not duplicate those bumps in this manual pin pass. The one thing that pass does own is the ignore entry for typescript majors in .github/dependabot.yaml: it suppresses the proposal Dependabot would otherwise make, so nothing else can surface it. Check npm view @astrojs/check peerDependencies.typescript against the current typescript major and delete the entry once the range covers it — the rule is version-agnostic, so left in place it blocks a major @astrojs/check fully supports just as silently as the TypeScript 7 it was added for (#3877).

• Runner images — every runs-on: label and matrix os:/runner: value in the workflows. A pin equals what its -latest label currently resolves to, per the availability table in actions/runner-images:

  gh api repos/actions/runner-images/contents/README.md \
    -H 'Accept: application/vnd.github.raw' | sed -n '/^| Image/,/^$/p'
  

  Bump any pin the table no longer lists against -latest, and update ci.yaml's header comment, which records the reason for each image it names. GitHub keeps two GA images per OS and begins deprecating the older one as soon as a newer goes GA, so a pin that has fallen off -latest is already the next one due; the table's deprecated badge marks that deadline, not the moment to move, and a row badged preview is not a bump target.

  A variant label — -arm, -intel, -large — has no -latest of its own; it follows its base image's row.

  Where a pin is deliberately held back, that comment says what the repo needs from the older image. Re-check that need against the newer image each week — upstream announces image changes as issues, so gh search issues --repo actions/runner-images "<the need> <newer image>" surfaces a reversal. Reading the cited issue's state is not the test — an announcement closes when its change ships, in either direction.

  The tend-*.yaml workflows also pin a runner, but tend's generator writes them and uvx tend@latest init overwrites a hand edit; file a tend issue for those.

Discovery shortcut: a recent green CI run on main flags cargo-install drift directly via workflow annotations. gh run view <run-id> --json jobs --jq '.jobs[].databaseId' | xargs -I{} gh api repos/<owner>/<repo>/check-runs/{}/annotations returns one warning per outdated pin.

Weekly Maintenance: Statusline Cache-Check

Detect new in-process cache-miss duplicates introduced by recent changes by profiling a real wt list statusline --format=claude-code trace. The render runs on every Claude Code prompt redraw, so duplicate git subprocesses there compound into measurable fseventsd / IPC load.

# Run from any worktree of this repo. `jq -n` builds the stdin JSON so the
# recipe is portable (the weekly job runs on ubuntu-24.04, whose GNU sed
# rejects BSD's `sed -i ''`) and so a path with a quote can't corrupt it.
jq -n --arg cwd "$PWD" '{
  hook_event_name: "Status",
  workspace: {current_dir: $cwd},
  model: {display_name: "Opus"},
  context_window: {used_percentage: 42.0}
}' > "${TMPDIR:-/tmp}/statusline-input.json"

# Debug build on purpose. `tend-weekly` installs no `wt` and restores no Rust
# cache, so `--release` means a cold optimized build of the whole dependency
# graph before the first render. The duplicate `(command, context)` pairs this
# check reads are profile-independent; only the timing columns, which this
# section doesn't triage, would be worth a release build.
cargo run -- -vv list statusline --format=claude-code \
  < "${TMPDIR:-/tmp}/statusline-input.json" > /dev/null
cargo run -- config state logs profile --format=json | jq .cache

The .cache report flags commands invoked more than once with the same context. Triage each duplicate:

• Legitimate (different cwd, different ref form that can't be normalized, intentional double-call across phases) — note in the response and move on.
• Cache miss (same logical operation should hit cache but doesn't) — open an issue or fix it. Common shapes: merge_base("main", "<sha>") vs merge_base("main", "branch") keying separately; worktree_at(cwd) vs worktree_at(porcelain_path) not canonicalizing.

Weekly Maintenance: LLM Model Names in Docs

Grep for current Claude and Codex pins across every tracked file:

git grep -niE "claude|codex"

Check the latest IDs at https://docs.anthropic.com/en/docs/about-claude/models and https://developers.openai.com/codex/models. The recommended commit-message commands should use the most recent fastest model from each vendor (Haiku for Anthropic, the smallest current Codex variant for OpenAI).

On drift, open a PR — don't file an issue. The source of truth is after_long_help in src/cli/mod.rs; edit it and let cargo test --test integration test_docs_are_in_sync regenerate the mirrors under docs/src/content/docs/ and skills/worktrunk/reference/. The "smallest current variant" call is a judgment — pick the one the vendor's models page currently positions as fastest/smallest, and explain the choice in the PR body. Verifying the new model name with an installed CLI (codex -m <name>, etc.) isn't possible in this CI sandbox; the PR is the right output anyway, and the maintainer tests on merge.

Weekly Maintenance: Agent App Integration Surfaces

Worktrunk ships a plugin for each agent CLI it integrates with, and those CLIs change their integration surfaces without notice. Each week, scan the upstream changelogs and flag changes that affect what Worktrunk consumes or produces.

| App | Source to check | Integration surface | |-----|-----------------|---------------------| | Claude Code | gh api repos/anthropics/claude-code/contents/CHANGELOG.md -H 'Accept: application/vnd.github.raw', plus curl -sL https://code.claude.com/docs/en/statusline.md for the statusline JSON schema | statusline stdin JSON, WorktreeCreate/WorktreeRemove hooks, plugin marketplace, /wt-switch-create | | Codex | gh release list -R openai/codex -L 10 | plugin marketplace | | Gemini CLI | gh release list -R google-gemini/gemini-cli -L 10 | native extension loading | | OpenCode | gh release list -R sst/opencode -L 10 | plugins API in ~/.config/opencode/plugins/ |

What to flag:

• New statusline JSON fields — src/commands/statusline.rs parses workspace.current_dir, model.display_name, context_window.used_percentage, and rate_limits.{five_hour,seven_day}.{used_percentage,resets_at}. A newly added field (session cost, PR review state) may be worth surfacing in wt list statusline.
• Renamed or removed hook events — WorktreeCreate/WorktreeRemove route agent worktree creation through wt; a renamed event silently disables isolation rather than erroring.
• Changed plugin install mechanisms — wt config plugins {claude,codex,opencode} install and the Gemini extension manifest break if the marketplace or plugins-directory contract changes.

Don't open a PR speculatively. File one issue per relevant change, linking the upstream entry and noting what Worktrunk would need to do. If nothing changed, say so and move on.

README Date Check

The README blockquote opens with a month+year (e.g., "April 2026"). During daily maintenance, verify the month matches the current month and update it if stale.

Per-Workflow References

• PR review: @references/review-pr.md — Rust idioms, documentation accuracy, duplication search
• Nightly sweep: @references/nightly-cleaner.md — survey checklist, branch naming, CI-breakage ownership