Plumbline 1.8.0

Changed

• Symptom-first skill descriptions. Lens skills now open with the symptoms users describe ("adding a config option is all ceremony", "one field means editing every file", "every new client gets it wrong", "PR review please") before the book's terms. red-flags defers to the matching lens when the user names a concern; diagnose triggers only when the user doesn't know what kind of problem they have. Measured across all 13 lens suites: the right skill now fires on 92% of positive cases (was 69%) and 92% of held-out cases (was 62%), negatives stay at 100%. See evals/TRIGGERS.md.
• Saying it's fine. diagnose and complexity-recognition lead with a clean verdict when the code shows no symptoms, instead of hunting for something to restructure (diagnose clean case 0.50 → 1.00).

Plumbline 1.7.0

Added

• Remembering findings. Optional .plumbline/findings.md in your project, keyed by concept. red-flags and design-review reconcile each review with it (new, still open, regressed, fixed) and update it; apply-fixes marks fixes; code-evolution flags regressions. Accepted entries with a reason count as ACCEPTED rationale.
• apply-fixes: coverage check, commit mode, flaky retry. Flags fixes no test reaches, can commit each fix separately (message names the flag), and re-runs once before reverting on an unrelated test failure.
• Design review on pull requests. Reusable GitHub Actions workflow (plumbline-review.yml) running code-evolution on a PR diff, commenting only on boundary/root-cause/regression findings. Read-only, forks skipped, prompt-injection guard, actions pinned to SHAs. See docs/ci-review.md.
• Docs site at https://tomymaritano.github.io/plumbline/, built from the skills on each release.
• Security and privacy section in the README.
• Design check for other agents. README snippet for AGENTS.md (and equivalents) that gives Codex, Cursor, Grok Build and other agents the review-on-stop behavior as an instruction, with the trade-off against the Claude Code hook spelled out.

Plumbline 1.6.0

Added

• Rust and Go guides for red-flags (references/rust.md, references/go.md): the idiomatic shapes each flag takes, such as single-impl traits and cfg-duplicated logic in Rust, producer-side interfaces and context-free return err chains in Go.
• An eval suite for every skill. The 13 skills that had none now have one (2 positive cases + 1 negative each, with baselines). New sandboxed suites: red-flags-scale (a ~2,000-line repo on disk with git history, measuring the reading strategy) and apply-fixes; see evals/MAC.md.

Plumbline 1.5.0

Added

• apply-fixes skill. Applies the findings of a red-flags or design-review report one fix at a time: checks for a clean tree, finds the test command, runs a baseline, shows a plan, asks before changing a public interface, runs the tests after every fix and reverts only the fix that broke them. Ends with applied / reverted / skipped and git diff --stat; never commits unless asked. red-flags, design-review and diagnose point to it.
• Review on stop (opt-in hook). With the review_on_stop option on, a turn that writes 80+ lines of code (review_threshold_lines) or creates a code file ends with one red-flags check of those files, reporting only what triggered. Off by default; Claude Code only. Dependency-free Node scripts in hooks/, tested with node:test (npm test) in CI.

Plumbline 1.4.0

Added

• Flag 18, Vocabulary Not Owned in red-flags: adapters (bindings, SDKs, clients) that each re-implement names, codes, parsing or defaults the core module should expose. New syndrome cluster in the flag interaction map, The Adapter Echo (Vocabulary Not Owned + Repetition + Information Leakage).
• ACCEPTED state in red-flags: a flag whose signal is present but backed by an explicit, still-valid rationale in the code or docs is listed in one line and left out of the summary and prioritization, instead of being reported as a problem.
• Reading strategy in red-flags for targets over about 1,000 lines or spanning several packages: public API first, then boundaries, then only the implementations a signal points at; tests and generated code skipped. design-review points to it for reviews at scale.
• Python and TypeScript guides (red-flags/references/python.md, typescript.md): the idiomatic shapes each flag takes in those languages.
• Read-only git access (git log, git diff, git show) for red-flags, code-evolution, information-hiding and design-review. code-evolution reviews a branch with git diff <base>...HEAD; the Git Co-Change Test now has a concrete command.
• Eval case 11-rust-bindings for red-flags, condensed from a real review of a Rust workspace with Node, Python and WASM bindings.
• Grok Build support. Install with grok plugin marketplace add tomymaritano/plumbline and grok plugin install plumbline. Native manifests live in .grok-plugin/ and are kept in sync by scripts/bump-version.sh.