Bump ruff floor to 0.16.0 and adopt its default rule set

[tony]

Summary

• Bump the ruff floor to >=0.16.0 in the dev and lint dependency groups so contributors and CI see the same diagnostics instead of splitting on whatever ruff each machine happened to resolve.
• Adopt ruff's own curated default rule set by moving [tool.ruff.lint] from select to extend-select. An explicit select replaces the default set rather than adding to it, so this project had been silently opting out of every rule it did not name itself. Enabled rules go from 351 to 565.
• Fix the six real findings that surfaced, one rule per commit, and scope the two intentional ones as per-file ignores with the reason recorded in pyproject.toml.

Why extend-select

ruff 0.16 ships a curated default rule set as its recommended baseline. There is no token that names that set from inside select — the selector grammar has ALL, prefixes, and individual codes, and nothing else — so a config with an explicit select cannot say "the defaults, plus these". Leaving select unset and layering the project's own linters on with extend-select is the only way to get defaults-plus-extras. The file carries a comment saying so, because the natural instinct on reading it is to "tidy" extend-select back into select.

Expanding to whole prefixes instead was considered and rejected: it drags in all of D, PL, S, and friends rather than ruff's curated subset, and is one to two orders of magnitude noisier for no corresponding gain in signal.

Findings fixed

FURB167 — re.S spelled out as re.DOTALL in the docs-drift test.

FURB188 — mcp_swap derived the server name with a conditional slice that restated the -mcp suffix twice and depended on -len() arithmetic staying in sync with its own endswith guard. Now str.removesuffix.

FURB192 — mcp_swap sorted a whole set of registered server names to read the first element, twice, both times to suggest a --server value. Now min.

FLY002 — the server's base instruction blob joined a static tuple of sections. ruff's own fix collapses the sections onto a single over-long line and trades the finding for an E501, so this one is hand-written instead: an f-string with one section per line, which reads the same as the tuple did.

ISC004 — the run_command marker-filtering fixtures build captured pane lines as multi-line implicit concatenations inside a list. Element boundaries are exactly what those fixtures test, and unparenthesized concatenation makes a dropped comma indistinguishable from a deliberate join. Parenthesized, matching the form one fixture in the same literal already used.

EXE001 — scripts/mcp_swap.py carried a shebang without the executable bit, and the interpreter it named was wrong besides: it is a PEP 723 script whose inline metadata declares tomlkit, so direct execution under bare python3 raised ModuleNotFoundError. The shebang now points at uv run --script, which honours that metadata, and the bit is set in the index. uv run scripts/mcp_swap.py and the just mcp-* recipes are unchanged. Worth flagging for anyone reproducing this locally: ruff does not enforce EXE001 on WSL, so this one surfaced only on CI.

Findings scoped as ignores

S102 in docs/conf.py — the Sphinx idiom of execing the package's __about__.py to read version metadata without importing the package into the docs build. A repo-local file, no user input. Scoped to that one file so a stray exec anywhere else still fails the lint.

BLE001 in six files — every site was read individually, and all of them are deliberate boundaries: the MCP error middleware that renders any tool exception as an is_error result, the per-operation isolation in the tool and send-keys batch runners, the best-effort tmux version probes and lifespan buffer GC, and mcp_swap's rollback to a config's original bytes when a write fails. tmux and the agent CLIs' config files fail in more ways than a narrowed clause can enumerate, so narrowing here would trade a logged degradation for a crash in code whose whole job is not to crash. BLE001 stays enforced everywhere else.

Design decisions

Floor, not pin. >=0.16.0 is what holds the version: it guarantees every contributor is at or past the release that changes formatting behavior, while still letting them take later patch releases. uv.lock pins the exact version CI resolves.

One commit per rule. Each fixed rule and each added ignore is its own commit, so a rule can be reverted or re-argued without unpicking the rest, and each commit body carries the rule's documentation link.

No noqa, no explanatory code comments. Intentional idioms are declared in the ruff config, where the next reader of the config finds them, rather than annotated at the call site.

Markdown formatting is a no-op here. 0.16.0 makes ruff format format Python and pycon code blocks inside Markdown by default, so the README and the docs/ tree are now in scope. Every Markdown file in this repo is already formatted the way ruff wants it, so ruff format --check . passes across the wider file set with no churn.

Test plan

• uv run ruff check . — all checks passed
• uv run ruff format --check . — every file already formatted (Python plus Markdown code blocks)
• uv run mypy — no issues found
• uv run pytest — full suite green
• Enabled-rule count confirmed against ruff check --show-settings . before and after the config change

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 86.08%. Comparing base (5c35acf) to head (64ca1cf).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #113   +/-   ##
=======================================
  Coverage   86.08%   86.08%           
=======================================
  Files          46       46           
  Lines        3593     3593           
  Branches      516      516           
=======================================
  Hits         3093     3093           
  Misses        355      355           
  Partials      145      145           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.