# Changes
## [0.4.0] — 2026-02-21
### Breaking
- `read_body()` now returns `Result<String>` instead of `String`, propagating
IO and parse errors instead of silently returning an empty string (#88)
### Added
- Symlink safety: new `fs_util` module with `is_regular_file()`,
`is_regular_dir()`, `is_symlink()` helpers; all file-system walks use
symlink-safe checks; `S005` diagnostic for symlinks in skill dirs (#87)
- Path traversal guard: `S006` diagnostic rejects `..` components in
reference links (#89)
- File size cap: `read_file_checked()` helper rejects files exceeding
1 MiB, preventing memory exhaustion (#90)
- Discovery error collection: `discover_skills_verbose()` and
`collect_skills_verbose()` collect warnings instead of silently
skipping unreadable paths; CLI commands print warnings to stderr (#91, #92)
- CRLF normalization in formatter: `format_content()` normalizes `\r\n`
to `\n` before byte-offset arithmetic (#94)
- Recursion depth limits (10 levels) in `copy_dir_recursive` and
`discover_skills_recursive` to prevent stack overflow (#96)
- Pre-tokenized Jaccard similarity in conflict detection eliminates
per-pair allocation in the O(n²) loop (#95)
- Formatter comment anchoring: standalone YAML comments stay in position
during key reordering instead of traveling with the preceding key (#103)
- Unified `scripts/version.sh` with `show`, `set`, and `bump` subcommands,
replacing `scripts/bump-version.sh` (#102)
### Fixed
- TOCTOU race in `build_skill()` and `init_skill()`: replaced
check-then-write with atomic `create_new(true)` (#93)
## [0.3.0] — 2026-02-21
### Breaking
- Renamed CLI subcommands: `validate` → `check`, `generate` → `new`,
`format` → `fmt`, `to-prompt` → `prompt` (#76)
- Old names (`validate`, `generate`, `format`, `lint`) retained as hidden
aliases for backward compatibility, but may be removed in a future release
- `tester` scoring formula changed to weighted combination:
name 15%, description 25%, trigger 20%, doc-comment 40% (#79)
- `CheckResult` now uses distinct `fail_label` for pass/fail display (#78)
### Added
- `build` subcommand: assemble one or more skills into a Claude Code plugin
artifact with `plugin.json`, reference files, and structured warnings (#83)
- `test` subcommand: fixture-based skill testing with YAML test suites,
`--generate` for auto-generating fixtures, and `--min-score` threshold (#84)
- `fmt` subcommand: canonical SKILL.md formatting — key reordering, trailing
whitespace removal, blank line collapsing, idempotent output (#76)
- `upgrade --full`: combined `--apply` + `--suggest` in a single pass (#80)
- `check` superset command: runs `validate` + `lint` together, replacing
separate invocations (#76)
- `AssembleWarning` public type for structured build warnings
- `scripts/bump-version.sh`: portable version bump across Cargo.toml,
plugin.json, CHANGES.md, and Cargo.lock (#82)
### Changed
- `upgrade --apply` uses YAML AST-preserving insertion for metadata blocks,
handling partial metadata, comments, and custom indentation correctly (#81)
- `BuildResult` includes structured `warnings` vec instead of `eprintln` (#45)
- `read_body()` deduplicated into `parser.rs` as single source of truth (#82)
### Fixed
- Path traversal validation in assembler (`is_unsafe_name`) rejects `..`
components and absolute paths
- JSON injection prevented: plugin.json generation uses `serde_json` instead
of string formatting
- `formatter` handles empty frontmatter without panicking
- `main.rs` exit codes: `test --generate` and `fmt` properly track errors
- `main.rs` parse error diagnostics shown in `check --no-validate`
- Delimiter matching uses `trim_end()` for robustness
- Doc-comment activation thresholds corrected to 0.4/0.15 in tester
- `bump-version.sh` uses portable `sedi()` wrapper for GNU/BSD sed
## [0.2.3] — 2026-02-20
### Added
- `score` command: rate skills 0–100 against a best-practices checklist
(structural 60pts + quality 40pts)
- `test` command: simulate skill activation with query matching
- `upgrade` command: detect and apply missing best-practice fields
(`compatibility`, `metadata.version`, `metadata.author`, trigger phrases)
- `doc` command: generate skill catalogs with diff-aware output
- `init` command: scaffold new skill directories from templates
- `lint` subcommand and `--lint` flag for semantic quality checks (I001–I005)
- `--structure` flag for directory structure validation (S001–S004)
- Cross-skill conflict detection (name, description, tool overlap)
- Watch mode (`--features watch`) for live re-validation on file changes
- JSON output format (`--format json`) for CI integration
- Claude Code plugin skill: `aigent-scorer`
- Comprehensive README with command examples and scoring documentation
### Changed
- Multi-skill prompt generation with `to-prompt` accepting multiple directories
- Builder supports building from existing SKILL.md files
### Fixed
- Jaccard similarity now case-insensitive (matching doc comment contract)
- `doc` catalog resolves `entry.location` to parent directory correctly
- `upgrade --apply` handles partial metadata blocks (inserts missing keys
under existing `metadata:` rather than skipping)
- Windows CI: platform-specific script permission checks (`#[cfg(unix)]`)
## [0.1.0] — 2026-02-20
### Added
- Core library: error types (`AigentError`), data model (`SkillProperties`),
YAML frontmatter parser, skill directory validator, XML prompt generator,
and skill builder
- CLI tool with subcommands: `validate`, `read-properties`, `to-prompt`,
`build`, `init`
- Full compliance with the
[Anthropic agent skill specification](https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices):
name/description validation, XML tag rejection, reserved word checks,
Unicode NFKC normalization, body-length warnings
- Dual-mode skill builder: deterministic (zero-config) and LLM-enhanced
(Anthropic, OpenAI, Google, Ollama) with per-function graceful fallback
- Claude Code plugin with two skills (`aigent-builder`, `aigent-validator`)
operating in hybrid CLI/prompt-only mode
- Install script (`install.sh`) for non-Rust users
- `--about` flag displaying project info from compile-time metadata
- Cross-platform support: Linux (x86_64, aarch64), macOS (x86_64, aarch64),
Windows (x86_64)
- CI pipeline: formatting, linting, testing, release builds on all platforms
- Release workflow: automated GitHub Releases and crates.io publishing on tag push