v1.1.3

• Test dir names come from test files only. A suite is named by
  files that look like tests (test_refund.py, refund.spec.ts,
  RefundTest.java). A README, fixture data, or contest.py no longer
  shows up as a tested feature.
• Generic names are dropped after the test marker is stripped.
  test_helpers.py no longer adds "helpers" to the cell.
• "test" is stripped only after a separator. contest_test.py
  gives "contest", not "con". A dotted name keeps its first part
  (AboutPanel.tabs.test.tsx gives "AboutPanel"), and names that differ
  only in case are listed once.

v1.1.2

• Sentence cuts respect abbreviations. A contains cell is cut at a
  sentence end only when ., !, or ? comes before a capital and
  does not close an abbreviation such as e.g., i.e., or etc..
• Test dirs keep repo-specific descriptions. Instead of a fixed
  phrase, an undocumented test dir is named by what its files test
  (test_refund.py, checkout.spec.ts give "tests for checkout,
  refund"). The repo's own docs still win.

v1.1.1

Scaffold fixes from a second real-repo pass. Generated maps for
KiroCrew, rysolv, ray-tracer, and RoBall3D now lint with 0 warnings.

• Submodules stay out of the index. A git submodule (gitlink, mode
  160000) is another repo and an empty dir in this one. It no longer
  counts as an uncovered top dir.
• The misc: row holds up to 24 dirs (was 12), so mid-size repos
  no longer push dirs into the unmapped: footer.
• Cleaner contains cells. Text is cut at the first sentence or at a
  word boundary with ..., never mid-word or mid-character. Section
  banners (== Foo ==) and commented-out code are skipped as
  descriptions. Test dirs read "tests and fixtures" instead of one
  test case's comment. C-family type names and include guards no
  longer show up as exports.

v1.1.0

Real-repo pass: the scaffold and search now hold up on 1,000 to 6,000
file codebases, not just fixtures. On a 6,000-file repo a broad
search.sh kiro: import dropped from about 338k tokens to under 1k.

• Search output is budgeted. search.sh prints at most 40 hits
  (--max N, --all, env PREFIXER_MAX_HITS), cuts lines past 240
  chars, and summarizes the rest by dir so the next call can narrow.
  -l lists files by hit count; -w and -F pass through to grep.
  The metrics line gains shown, and est_tokens is what printed.
• Git-aware file index. lib_scope.sh lists files once per run
  from git ls-files (tracked plus untracked-not-ignored), falling
  back to find outside git. Scopes, widen-to-root, and coverage
  checks all read it, so .gitignored output never reaches a search
  and no scope runs its own find.
• Size-weighted scaffold. init_namespaces.sh weighs every dir by
  the text files a grep would read, then gives rows to the heaviest
  dirs up to three levels deep. A big package gets child rows instead
  of one broad scope, single-child chains collapse (src/pkg/ maps as
  pkg:), and container dirs (src/, services/, packages/) keep
  their children as top-level tags.
• No silent coverage holes. Top dirs that miss the row cap share
  one misc: row; the unmapped: footer only lists what did not fit
  there, capped at 8 names.
• Better contains cells. Each row reports its file count and the
  biggest subdirs without a row. Descriptions also come from the dir's
  own README or manifest, agent-doc lines stop at the next path, and
  symbols are sampled from source files only.
• Tighter synthetic rows. Root files compress to *.ext groups,
  lockfiles stay out, and jumbo-file rows only name source files that
  are not tests.

v1.0.0

First stable release. The skill, scripts, spec, and eval harness from
0.x carry over unchanged; 1.0 adds the release machinery, community
files, spec conformance, and a one-command health check a public
project needs.

• scripts/doctor.sh: one-command health check. Reports the map,
  lint verdict, session-start hook state, metrics gitignore coverage,
  and stale install stamps (.prefixer_version vs the running
  checkout). ok/warn/FAIL lines, nonzero exit on failures,
  --quiet for CI. SUPPORT.md and the bug-report form now ask for its
  output.
• scripts/bump-version.py: one-command release bump. Rewrites the
  version in SKILL.md, all three plugin manifests, docs/index.html's
  softwareVersion, and the integrations/ CI pin, then opens a dated
  CHANGELOG section. RELEASING.md documents the full process, and
  release.yml creates the GitHub release when a v* tag lands,
  gated on validation, tests, and real changelog notes.
• agentskills.io spec conformance. Added the compatibility field
  (bash + POSIX tools; Python 3 for visualize.py) and dropped the
  triggers field: the official validator rejects non-spec
  frontmatter, and [user, model] is Devin's default anyway, so the
  field was redundant. The official skills-ref validator now runs in
  CI pinned at skills-ref==0.1.1.
• install.sh gains a windsurf platform. Project
  .windsurf/skills/prefixer, global ~/.codeium/windsurf/skills/prefixer,
  matching the paths Devin's docs publish. Detection markers:
  .windsurf/ and ~/.codeium/windsurf/.
• Community files. CODE_OF_CONDUCT.md (Contributor Covenant),
  SUPPORT.md, ROADMAP.md (the 1.0 tracker), and a
  .pre-commit-config.yaml that runs the CI gates on commit.
• summarize.py honest phrasing. A negative byte delta now reads
  "N% larger" instead of "-N% smaller"; RESULTS.md regenerated.
• validate-package.py gains a check: the consumer workflow's
  script-fetch pin must equal the release version.
• run.sh --dry-run works with no agent CLI. Adapter picking used
  to require a detected binary even though a dry run never calls one;
  machines with no agent CLI (like CI) fell over at pick_adapter. The
  auto path now falls back to the devin adapter so hook wiring can be
  verified anywhere, and the not-ready message covers missing binaries
  as well as missing logins.
• README, docs site, and llms.txt synced: windsurf install, doctor.sh,
  and the agent-eval aggregate numbers (4.6 vs 7.2 mean tool calls per
  task, mapped vs baseline).

v0.7.1

SEO and agent-discovery pass.

• SKILL.md description gains "save tokens" and "cut context cost" trigger phrases
• Landing page: hook line, FAQ section with FAQPage JSON-LD, agent-facing install callout
• llms.txt for LLM-facing site description
• README agent callout block
• Repo description now leads with "save tokens"; topics updated

v0.7.0 — measured token savings

What's new

• eval_repo.sh — lints a map, then probes every namespace with its own name as the grep pattern. Zero per-repo config; prints a per-row hit/byte table and mean pct_saved.
• check_namespaces.sh — map lint: dead scopes, duplicate tags, draft rows, generated dirs, coverage gaps.
• eval_scope.sh — single-query token delta: root grep vs scoped grep.
• Scaffold derives prefixes from the codebase's own vocabulary, nests sub-team rows (payments:checkout:), emits repo: slugs, and lands zero-TODO drafts.

Measured on public repos

repo	namespaces	mean bytes saved	range
obra/superpowers	8	89%	69–100%
DietrichGebert/ponytail	12	91%	68–100%

Install

npx skills add Albisourous/prefixer — or copy SKILL.md, references/, scripts/ into .agents/skills/prefixer/.