v1.23.1

Fixes

• catch reference-only skill changes in update/doctor, and tidy what the audit found (#17) (7264fec)

Documentation

• regenerate CHANGELOG.md for v1.23.0 (ed5fa38)

Maintenance

• bump version to 1.23.1 (eef-install update/doctor catch reference-only changes, skill-tree installs drop stale files) (ee50a3e)

v1.23.0

Features

• security-review: boundary-and-result gate and a Needs-validation verdict (#12) (cf182a7)
• security-review: Mode C, a lean full-repo audit before production (#13) (d784916)
• security-review: eight boundary references from Cloudflare's domain companions (#14) (88f411f)

Documentation

• regenerate CHANGELOG.md for v1.22.1 (9fdec29)

Maintenance

• bump version to 1.23.0 (security-review: boundary gate, Needs-validation verdict, Mode C full audit, eight boundary references) (0f22898)

v1.22.1

Fixes

• release: stop the release flow tripping over its own changelog (#9) (6d045e5)

Documentation

• regenerate CHANGELOG.md for v1.22.0 (c8bd453)

Maintenance

• bump version to 1.22.1 (release flow: changelog check on main pushes, wait for the right publish run) (10c88bd)

v1.22.0

Features

• cut agent quota burn, guard destructive commands, show usage (#4) (474bfa2)
• Windows CI, one generator list, review dates, outcome evals, incident response (#5) (6d765b9)
• use Salak 0.3.x across skills, rules and doctor (#6) (4ce234a)

Fixes

• scripts: skip deleting a release branch gh already removed (0e7c566)

Documentation

• regenerate CHANGELOG.md for v1.21.0 (d3fdafe)

CI and release

• publish: publish through OIDC only now that the npm token is gone (#3) (bc2f199)

Maintenance

• bump version to 1.22.0 (Windows CI, one generator list, shorter skill descriptions, outcome evals, incident response, Salak 0.3.x) (487296b)

v1.21.0

Features

• scripts: cut a release in one command with release.py (cb388fc)
• evals: add light-tier benchmark for research-worker and docs-sync (3143568)
• evals: make benchmark records append-only and land work through PRs (c861edd)
• scripts: land release bump and changelog through PRs (bbc4a70)

Fixes

• scripts: write generated files with LF on every platform (b8d6708)
• scripts: read release.py child output as UTF-8 on Windows (c94866d)
• skills: make language-code-review load for a framework review of a few files (c32cb2f)
• agents: let the haiku alias take effort now that it is Haiku 5.5 (91976b1)
• skills: make performance-audit load when reported slowness comes with the code (0c8c5a2)

Documentation

• regenerate CHANGELOG.md for v1.20.1 (e109a96)
• evals: record the light-tier run - keep research-worker and docs-sync light (28e14ef)

Tests and evals

• evals: cover all 38 skills with trigger cases and near-miss checks (bec585d)
• evals: record the coverage trigger run (spend limit hit) (3a82ac8)
• evals: record coverage re-runs and the three description fixes (eaf60e1)
• evals: record run 7 — performance-audit fires, seo-audit inline stays a known miss (d2567a2)

Maintenance

• bump version to 1.21.0 (tier routing benchmarks, append-only records, PR-based releases) (5540918)

v1.20.1

Features

• code-review: escalate the reviewer on logic changes, not file count (D-062) (ace8d41)
• evals: add a v4 corpus of 22-26-file logic PRs to test D-062 (2420fb5)

Documentation

• regenerate CHANGELOG.md for v1.20.0 (f415aae)
• evals: record the v4 run — first tier difference, D-062 holds (8eed781)

Maintenance

• bump version to 1.20.1 (reviewer escalation on logic changes, D-062) (02ad356)

v1.20.0

Features

• install: add eef-install update for stale Claude Code installs (3c8d1ec)
• scripts: generate CHANGELOG.md and release notes from commits (8b79054)
• evals: add a v3 corpus of large PRs for the reviewer threshold (2f98eb1)

Fixes

• skills: make code-review and code-simplification load for quick requests (5f27ab9)
• skills: make build-fix load for a single pasted build error (9e23bb7)
• agents: verify Codex models against Codex's own catalog (ee9fd83)
• evals: remove real bugs from the v3 base project, score PR files only (a78ba36)
• evals: restore the full v1 benchmark scores (e364fbc)

Performance

• skills: trim the five largest SKILL.md files (~6.6k tokens across them) (7fa2ff3)
• code-review: keep PR Review Mode inline, compacted (e072dc1)

Documentation

• evals: record the v3 large-PR benchmark run (78cad77)
• evals: correct findings-per-case figures in RESULTS-v3 (2825400)

Tests and evals

• evals: record review benchmark after the SKILL.md trim — no regression (5bb49f8)

Maintenance

• bump version to 1.20.0 (eef-install update, changelog, trigger fixes, large-PR benchmark) (5f71e63)

v1.19.0

Model routing by tier (D-060, D-061)

• Every agent declares a harness-neutral tier: (light / standard / deep) and effort: instead of a fixed model: sonnet. The tier follows one question: is a wrong answer from this agent loud or quiet?
• agents/model-profiles.json maps tiers to models per harness. Pick a cost profile at install: --profile hemat | seimbang (default) | maksimal.
• --models <file> sets your own model ID per tier — required for OpenCode/ZCode, and an override for Claude Code/Codex when a model is renamed.
• code-reviewer now runs at standard (Sonnet): on a seeded-bug benchmark every tier caught 27/27 bugs (single-file and context-dependent multi-file) at about half Opus's cost. For large PRs the code-review skill tells the caller to run the reviewer one tier up. code-critic and security-review stay deep.

Codex

• New target eef-install --target codex-agents (--global for ~/.codex/agents): 46 custom-agent TOML files, read-only sandbox for review/audit agents.

Installer

• eef-install --target claude-hooks installs the fs-guard and subagent-telemetry hooks and merges them into settings.json without touching other keys.
• eef-install doctor reports what is installed, what is stale, and which model profile is active.

Skills

• Orchestrator skills (research-ops, click-path-audit, spec-mining, gan-harness) escalate one failing delegate one tier up instead of the whole fan-out.
• config-hygiene documents the advisor tool and ships SubagentStop telemetry plus a report for re-tiering agents from evidence.
• git-and-release-ops now loads for quick git questions (it was not firing).

Lighter sessions

• Agent descriptions no longer restate their skill's description: per-session context drops from ~16.7k to ~12.2k estimated tokens. CI caps it at 13.5k.

Evals

• evals/review-seeded-bugs/ — seeded-bug benchmark for review tiers (v1 single-file, v2 multi-file).
• evals/triggers/ — claude plugin eval cases checking natural (mostly Indonesian) requests reach the right skill.

Release pipeline

• npm publishing moves to trusted publishing (OIDC) and runs the full validation and export-sync checks before npm publish.
• Line endings normalized to LF via .gitattributes.

v1.18.0

New skills (35 → 38)

• counterparty-comms-edho-ferdian — rules for agents in channels that outside parties read (audience and mention gating, no internal leakage), plus a hash-bound operator approval loop with a tested SQLite reference ledger.
• legal-ops-edho-ferdian — reproducible agreement drafts from a template plus a JSON spec (always DRAFT, not legal advice), a Schedule A notice procedure, and gated e-signature envelope preparation. Ships a dependency-free build-agreement.js.
• video-style-edho-ferdian — measure reference footage into a style pack (zone chroma LUT, cut cadence, overlay plates), then grade, cut and numerically verify new footage against it.
• backend-engineering-edho-ferdian gains references/rails.md (Rails 7.1+/8.x).

Always-on rules

• New rules/ (4 short files) and installer target npx eef-install --target claude-rules → ~/.claude/rules/eef/.

Fixes

• Agent wrappers now preload their wrapped skill (skills: frontmatter on Claude Code) and ban filesystem-wide searches. Previously a delegated agent could run find / -name SKILL.md, which on Windows scanned the whole drive for hours.
• safe-execution-edho-ferdian ships hooks/block-fs-wide-search.js, a PreToolUse hook that denies recursive searches from a filesystem, drive or home root.
• The validator now enforces agent skills: fields, bans root-level searches, and checks rule pointers. The pre-commit hook now also runs on agents/, scripts/ and rules/ changes.

Agents: 43 → 46. Reinstall with npx eef-install, --target claude-agents, and --target claude-rules.

v1.17.4

Supersedes v1.17.3, which failed to publish to npm (prepare script
required husky in an environment without devDependencies installed —
fixed here, see commit c243fd5).

Skill-hygiene pass

• Add the §10 "Surgical changes" canonical-contract pointer (or explicit carve-out) to frontend-engineering-edho-ferdian, backend-engineering-edho-ferdian, dead-code-cleanup-edho-ferdian, code-simplification-edho-ferdian
• Trim 10 bloated description: frontmatter fields (stacks/modes/lenses/drivers moved to each skill's body) — dev-kickoff-edho-ferdian was 2 chars from the hard 1024-char CI gate
• Fix stale "33 skills" → 35 in SPONSORS.md and skill-authoring-edho-ferdian; annotate (without rewriting) the historical 33/34-skill counts in README's point-in-time audit narrative
• Regenerate all cross-harness exports (AGENTS.md, CATALOG.md, dist/*.skill, .cursor/rules, .windsurf/rules, .devin/rules, .clinerules, .github/copilot-instructions.md, .kiro/skills, GEMINI.md, .hermes/skills, .agents/skills, .zcode/skills) to match

Tooling

• Add a .husky/pre-commit hook that locally mirrors CI's validate/dist-sync/export-targets-sync jobs whenever skills/ changes, so this class of drift is caught before it's committed

No skill behavior changes — content reorganization and hygiene only.