is-it-fixable

Is this GitHub issue actually fixable and worth your time — or a trap?

Most "find an issue" tools tell you whether an issue is claimed. is-it-fixable answers the harder question every contributor actually cares about: should I pick this one? It scores each open issue and shows you the evidence behind the verdict — so you don't fork a repo, set up its toolchain, and read the code only to discover the bug was already fixed, the PR already exists, or a maintainer already said "won't fix."

Why

Existing discovery tools (goodfirstissue.dev, up-for-grabs, and even claim-checkers) stop at "is someone working on it?". They don't tell you an issue is a dead end because:

• a closed, unmerged PR already tried and got rejected,
• a maintainer commented "won't fix" / "by design" / "out of scope",
• it was reopened after a PR as a staged/blocked change,
• it's tagged wontfix / needs-discussion / duplicate,
• or it may already be fixed on the default branch.

is-it-fixable scores all of that into a single verdict with transparent evidence.

Install

npm install -g is-it-fixable
# or run without installing:
npx is-it-fixable <owner/repo>

Usage

# Triage the open issues of a repo (most-fixable first)
is-it-fixable toml-rs/toml

# A single issue
is-it-fixable https://github.com/toml-rs/toml/issues/1120
is-it-fixable toml-rs/toml#1120

# Machine-readable output for scripts / CI
is-it-fixable apache/commons-lang --json

# Also report whether YOU can build it locally
is-it-fixable smithy-lang/smithy --build

# For a repo target, a merge-velocity signal is shown by default (disable with --no-repo-health)
is-it-fixable toml-rs/toml --no-repo-health

# Raise the GitHub rate limit (60/hr -> 5000/hr)
export GITHUB_TOKEN=ghp_xxx        # or: --token ghp_xxx

Tip: a GITHUB_TOKEN is optional but strongly recommended — unauthenticated GitHub API calls are capped at 60/hr. Any classic token with default (public) scope works.

Example

is-it-fixable — example/mixed-repo
4 open issue(s) analyzed · 1 look CLEAN

🟢  CLEAN  (score 65/100)  #7 Typo in error message for invalid config key
    https://github.com/acme/widgets/issues/7
    • [+15] Contribution-friendly label(s): good first issue, bug.

🔴  TAKEN  (score 5/100)  #6136 @JsonFilter with serializeAllExcept() does not filter @JsonAnyGetter entries
    https://github.com/FasterXML/jackson-databind/issues/6136
    • [-45] Open PR #6134 is already linked (someone is fixing this).

🟠  CONTENTIOUS  (score 0/100)  #1120 Error highlight doesn't handle multi-byte characters correctly
    https://github.com/toml-rs/toml/issues/1120
    • [-25] Closed, unmerged PR(s) #1132, #1139 — a prior fix attempt was rejected or abandoned.
    • [-25] A maintainer comment signals this may not be fixed (e.g. won't-fix / by-design / out-of-scope).

Verdicts

| Verdict | Meaning | |---|---| | 🟢 CLEAN | No claim/contention signals — genuinely available and worth a look. | | 🟡 STARTED | A draft PR or linked branch exists — proceed carefully. | | 🔴 TAKEN | An open (non-draft) PR is already linked. | | 🟠 CONTENTIOUS | Closed-unmerged PR, downgrade label, reopened-after-PR, or a maintainer "won't fix". | | ⚪ STALE | Long-idle with no maintainer engagement. |

Each verdict comes with a 0–100 score and a list of evidence — every signal cites the data that produced it, so nothing is a black box.

Exit codes

• 0 — at least one CLEAN issue was found (there's something to pick).
• 1 — no CLEAN issue found.
• 2 — an error occurred (bad target, rate limit, auth, not found).

How it works

is-it-fixable uses the GitHub GraphQL API to read each issue's timeline (cross-referenced & connected PRs), labels, comments (with author association), reactions, linked branches, and reopen history — then runs a deterministic, explainable rubric (no LLM). It also searches for pull requests that reference the issue (type:pr), because many real fix PRs never generate a timeline cross-reference — so an issue that already has an open fix PR is correctly reported TAKEN rather than CLEAN. Maintainer sentiment is trusted only from OWNER / MEMBER / COLLABORATOR comments, so a random user arguing "this is by design" doesn't downgrade an issue.

For a single-issue check it adds two "look before you leap" signals learned from real misfires: error-signature cross-search (extracts the issue's panic/assertion/exception/source-location signature and searches the repo for other issues/PRs carrying it, catching a bug already known or attempted under a different number), and a repo openness / merge-velocity warning (LOW-EXTERNAL-MERGE when a repo rarely merges outside PRs, CLOSED-TO-EXTERNAL when it appears to restrict PR creation to collaborators — bots excluded), so you don't craft a fix a repo won't take.

It also detects prose-only PR references — a PR mentioned merely in a comment or the issue body (#1195, or a …/pull/1195 URL) creates no structured timeline event, yet often marks a prior rejected attempt. These are weighted more cautiously than structural links: a mentioned closed-unmerged PR flags CONTENTIOUS, while a mentioned open PR is a moderate downgrade rather than a hard TAKEN.

For a repo target it also reports a merge-velocity signal — how readily the repo merges external (non-member) PRs (HEALTHY / SLOW / LOW-EXTERNAL-MERGE / UNKNOWN, with counts and median time-to-merge). This is a repo-level signal only: it never changes a per-issue verdict, so a genuinely CLEAN issue is never hidden just because its repo is slow — it just tells you whether your effort is likely to be reviewed at all.

Not affiliated with GitHub. is-it-fixable is a heuristic aid, not an oracle — always sanity-check before committing hours. "Already fixed on the default branch" detection (build + reproduce) is planned for a future release.

Contributing

Contributions are very welcome — see CONTRIBUTING.md. The rubric lives in src/rubric.ts and every signal is covered by tests, including an accuracy gate of real, hand-graded issues.

License

Apache-2.0 © Nagendra Mohan