Skip to content

CHANGELOG.md

Latest commit

 

History

History
201 lines (156 loc) · 14.3 KB

CHANGELOG.md

File metadata and controls

201 lines (156 loc) · 14.3 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

Added

Changed

Fixed

2.1.1 - 2026-06-09

Changed

  • Dependency updates (Renovate); CI now uses org reusable workflows.

Fixed

  • Suppress always-true PHPStan narrowing on the Composer tryComposer guard.

2.1.0 - 2026-05-12

Added

  • composer outdated + direct skills — after Composer’s package table (text mode), lists direct skills whose lock pin lags git ls-remote or local path content; JSON mode keeps package JSON valid and points to composer skills:outdated -f json. New command composer skills:outdated (--strict for CI exit 1).
  • Semver constraints on GitHub refs — owner/repo:^1.2 or --ref='^1.2' resolves to the highest remote tag matching the constraint (composer/semver + git ls-remote --tags); composer update re-resolves and refreshes composer.skills.lock while composer install keeps the pinned commit.
  • Direct skill installation — declare sources under extra.ai-agent-skills, pin them in composer.skills.lock, and materialize trees under install-dir (default vendor/agent-skills/installed/). CLI: composer skills dispatcher with skills:add, skills:install, skills:update, skills:remove, skills:list (see README, docs/IMPLEMENTATION-DIRECT-SKILLS.md, ADRs 009–012).
  • Git cache directory — ephemeral clones and worktrees use cache-dir (default vendor/agent-skills/cache). Config keys install-dir, sources-dir, and cache-dir plus lock path fields are validated (no .., no absolutes) so installs cannot escape the project root on a tampered lock.
  • Shared trust store for direct skills — allow/deny uses extra.ai-agent-skill.allow-skills with keys direct:<source>/<skill-name>; composer skills:trust / list-skills / read-skill behave like package skills.
  • DiscoveredSkills helper — merges package + direct discovery for list-skills / read-skill; duplicate names within Composer packages keep the first and print a [NOTE], while a package vs direct name collision now errors (aligned with SkillPlugin / AGENTS.md generation).
  • FilesystemUtil — 0755 directory mode, shared recursive tree removal with optional verbose IO diagnostics (replaces duplicated rmTree / 0777 usage).

Changed

Fixed

  • Direct skills security hardening — reject symlinks when copying and hashing skill trees; validate composer.skills.lock git commits as 40-char hex before using them as cache path segments; allow path-source url: "." in the lock; validate clone/tag refs so they cannot start with - (git CLI ambiguity). Note: git checkout -- <rev> is not used because Git treats a lone argument after -- as a pathspec, not a revision.
  • PR review follow-ups — GitHub HTTPS repo URLs may contain dots in the name (regex delimiter fix); strict project-root prefix for stored path sources (no sibling-directory false positives); skills:outdated validates lock url / path / install-path like install does so a tampered lock cannot drive reads outside the project.

2.0.0 - 2026-05-01

Breaking changes

First-run trust default changed from "auto-trust everything" to "deny by default". If you upgrade an existing installation that was implicitly relying on the v0.1.x behaviour where every type: ai-agent-skill package was auto-registered in AGENTS.md, you will now see a one-time prompt asking how to seed the trust map ([n] None / [d] Direct deps only / [a] All). Choose [a] to preserve the previous behaviour. Non-interactive runs (CI) default to [n] with a composer skills:trust … recovery line per affected package.

See #42 and the security review on PR #43 for the rationale.

Added

  • PHP 8.5 and Symfony 8.0 support in the test matrix and Composer constraints (symfony/yaml, symfony/console).
  • Symfony 7.4 LTS added to the test matrix (replacing 7.2, which no longer receives security updates).
  • Composer host matrix — every PHP × Symfony combination is now tested against both Composer 2.2 LTS and 2.9 (the only supported Composer 2.x release lines).
  • Lowest declared dependency validation — one extra row resolves with --prefer-lowest --prefer-stable against Composer 2.2 LTS, ensuring documented minimums actually install and pass tests.
  • Universal skill discovery: any Composer package can now ship skills via extra.ai-agent-skill, regardless of its declared type. Closes #42.
  • Trust prompt: first-time discovery from a new package prompts the user (y/n/a/d) before registering its skills. Decisions persist in root composer.json under extra.ai-agent-skill.allow-skills with glob support, mirroring Composer's config.allow-plugins.
  • First-run policy prompt for legacy type: ai-agent-skill packages: [n] None / [d] Direct deps only / [a] All, default n (strict). Non-interactive mode defaults to n with a per-package composer skills:trust ... recovery hint, so CI never silently auto-trusts dependencies. Replaces the earlier prototype's "auto-seed everything" behavior flagged HIGH by the security review.
  • Root package is auto-trusted — projects that ship their own SKILL.md no longer get prompted to authorize themselves.
  • composer list-skills now shows trust state ([allowed] / [pending] / [denied]) per skill and a footer count of pending packages. The command is purely informational and never prompts.
  • composer skills:trust <package> command — allow (composer skills:trust vendor/foo), deny (--deny), or revoke (--revoke) a trust decision without hand-editing composer.json. Used as the recovery path for accidental denies and as the canonical fix for non-interactive failures.
  • composer skills:list-trust command — read-only inventory of every persisted decision in extra.ai-agent-skill.allow-skills, with [allowed] / [denied] and (exact) / (glob) markers per entry. Companion to skills:trust; never prompts, never mutates.
  • Trust prompt mirrors Composer's plugin prompt shape (y/n/a/d/?). The ? answer shows per-option help and re-prompts. The prompt also includes an inline composer skills:trust <package> recovery hint so users have an on-screen breadcrumb if they pick n by accident.
  • composer read-skill now shows trust state in the header and warns when reading content from a pending or denied skill (which is not registered in AGENTS.md).
  • SkillTrustManager, PackageProvider, InstalledVersionsProvider, PackageInfo, and TrustDecision abstractions for testability.

Fixed

  • Composer 2.2 LTS compatibility — CommandContextTrait now falls back to the legacy getComposer(false) API on Composer 2.2; tryComposer() was only introduced in Composer 2.3. Detected by the new --prefer-lowest matrix row.

Changed

  • composer-plugin-api constraint changed from ^2.1 to 2.2.*|^2.9 — matches the only Composer release lines we test and the only ones that receive upstream support. Composer 2.0/2.1 are out of support; 2.3–2.8 are no longer maintained either.
  • composer/composer (require-dev) constraint changed from ^2.1 to 2.2.*|^2.9 for the same reason.
  • SkillDiscovery no longer filters by package type. Legacy type: ai-agent-skill packages with a root SKILL.md continue to work unchanged.
  • SkillDiscovery::discoverAllSkills() is now pure — it enumerates every declared skill with a trust_state field but never prompts. Gating happens at the install/update boundary in SkillPlugin::updateAgentsMd() only.
  • Non-interactive composer install now skips untrusted skill packages with a composer config --json hint instead of registering them silently.
  • PHPStan level bumped from 8 to 10 (max).

1.1.5 - 2026-04-20

Changed

  • Dependency updates via Renovate: step-security/harden-runner v2.17.0–2.19.0, actions/cache digest refresh, dependabot/fetch-metadata v3.x, codecov/codecov-action v6.

1.1.4 - 2026-03-20

Security

  • GitHub Actions hardening (#31): SHA-pin all third-party actions and add Dependabot for the github-actions ecosystem so action updates ship as reviewable PRs rather than floating tags.

Changed

  • Dependency updates via Renovate: step-security/harden-runner v2.15.0–2.16.0, shivammathur/setup-php digest refresh, codecov/codecov-action digest refresh.

1.1.3 - 2026-02-09

Added

  • "Agent Skills" branding and cross-platform compatibility (#10).
  • Auto-merge workflow for vetted dependency updates.
  • Renovate configuration (#1).

Security

  • Pinned GitHub Actions to commit SHAs with explicit per-job permissions (#2).

Changed

  • PHPUnit upgraded to v13 (#21) — breaking-change adaptations in the test suite.
  • Multiple Renovate-driven action and digest updates: step-security/harden-runner v2.14.0–2.14.2, actions/checkout v6.0.2, actions/cache digest refreshes, dependabot/fetch-metadata v2.5.0.

1.1.2 - 2025-11-26

Fixed

  • Skills Section Clarity: Clarified that project skills supplement Claude Code's built-in Skill tool capabilities
    • Changed ambiguous "Only use skills listed" to "For project-specific tasks, only use skills listed"
    • Added note explaining native capabilities remain available alongside project skills
    • Prevents AI agents from ignoring their built-in skill system

1.1.1 - 2025-11-25

Added

  • Working Directory Reminder: read-skill command now displays actionable footer with copy-paste ready cd command to help AI agents execute scripts from correct directory

Changed

  • AGENTS.md Instructions: Updated base directory instruction from descriptive to imperative language for clearer guidance

1.1.0 - 2025-11-25

Added

  • Symfony 5.4 LTS Support: Extended compatibility to support Symfony 5.4+ (previously 6.0+)
    • Now supports: Symfony ^5.4|^6.0|^7.0
    • Enables usage in projects still on Symfony 5.4 LTS
  • GitHub Actions CI: Comprehensive continuous integration workflow
    • Test matrix across PHP 8.2, 8.3, 8.4
    • Test matrix across Symfony 5.4, 6.4, 7.1
    • Lowest dependencies testing (PHP 8.2 + Symfony 5.4)
    • Code quality checks (PHPStan level 8, PHP-CS-Fixer)
    • Code coverage reporting with Codecov integration
    • Automated testing on every push and pull request

Changed

  • Library Best Practices: Removed composer.lock from repository
    • Libraries should not commit lock files
    • Added composer.lock to .gitignore
    • Ensures proper dependency resolution for consumers

Fixed

  • Installation Documentation: Updated README with accurate Composer 2.2+ plugin authorization requirements
    • Added interactive installation prompt example
    • Added non-interactive/CI installation instructions
    • Documented allow-plugins configuration requirement

1.0.0 - 2025-11-24

Added

  • Automatic Skill Discovery: Discovers all installed packages with type ai-agent-skill
  • AGENTS.md Generation: Generates openskills-compatible XML skill registry
  • CLI Commands:
    • composer list-skills - List all available AI agent skills with package info
    • composer read-skill <name> - Display full SKILL.md content for a specific skill
  • SKILL.md Parsing: Validates and extracts YAML frontmatter following Claude Code specification
  • Multiple Skill Support: Packages can provide single or multiple skills via configuration
  • Convention Over Configuration: Zero-config setup with SKILL.md in package root
  • Comprehensive Validation:
    • Name format validation (kebab-case, max 64 chars)
    • Description length validation (max 1024 chars)
    • YAML syntax validation
    • Required field validation (name, description)
  • Edge Case Handling:
    • Duplicate skill names (last wins with warning)
    • Invalid frontmatter (skip with warning)
    • Missing SKILL.md files (skip with warning)
    • Malformed YAML (skip with detailed error)
  • Configuration Options:
    • extra.ai-agent-skill for custom skill paths
    • Support for single skill (string) or multiple skills (array)
  • Base Directory Support: Outputs directory containing SKILL.md for resource path resolution
  • Progressive Disclosure: Lightweight XML index, full details on demand
  • Atomic File Updates: Safe AGENTS.md updates with temp file + rename
  • Content Preservation: Updates only <skills_system> block, preserves other AGENTS.md content

Security

  • Absolute Path Rejection: Blocks absolute paths in extra.ai-agent-skill configuration
  • Path Traversal Protection: Uses realpath() to resolve canonical paths
  • XML Escaping: Properly escapes skill metadata in generated XML