feat(extensions): Quality of life improvements for RFC-aligned catalog integration (#1776)

* feat(extensions): implement automatic updates with atomic backup/restore

- Implement automatic extension updates with download from catalog
- Add comprehensive backup/restore mechanism for failed updates:
  - Backup registry entry before update
  - Backup extension directory
  - Backup command files for all AI agents
  - Backup hooks from extensions.yml
- Add extension ID verification after install
- Add KeyboardInterrupt handling to allow clean cancellation
- Fix enable/disable to preserve installed_at timestamp by using
  direct registry manipulation instead of registry.add()
- Add rollback on any update failure with command file,
  hook, and registry restoration

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(extensions): comprehensive name resolution and error handling improvements

- Add shared _resolve_installed_extension helper for ID/display name resolution
  with proper ambiguous name handling (shows table of matches)
- Add _resolve_catalog_extension helper for catalog lookups by ID or display name
- Update enable/disable/update/remove commands to use name resolution helpers
- Fix extension_info to handle catalog errors gracefully:
  - Fallback to local installed info when catalog unavailable
  - Distinguish "catalog unavailable" from "not found in catalog"
  - Support display name lookup for both installed and catalog extensions
- Use resolved display names in all status messages for consistency
- Extract _print_extension_info helper for DRY catalog info printing

Addresses reviewer feedback:
- Ambiguous name handling in enable/disable/update
- Catalog error fallback for installed extensions
- UX message clarity (catalog unavailable vs not found)
- Resolved ID in status messages

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(extensions): properly detect ambiguous names in extension_info

The extension_info command was breaking on the first name match without
checking for ambiguity. This fix separates ID matching from name matching
and checks for ambiguity before selecting a match, consistent with the
_resolve_installed_extension() helper used by other commands.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* refactor(extensions): add public update() method to ExtensionRegistry

Add a proper public API for updating registry metadata while preserving
installed_at timestamp, instead of directly mutating internal registry
data and calling private _save() method.

Changes:
- Add ExtensionRegistry.update() method that preserves installed_at
- Update enable/disable commands to use registry.update()
- Update rollback logic to use registry.update()

This decouples the CLI from registry internals and maintains proper
encapsulation.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(extensions): safely access optional author field in extension_info

ExtensionManifest doesn't expose an author property - the author field
is optional in extension.yml and stored in data["extension"]["author"].
Use safe dict access to avoid AttributeError.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(extensions): address multiple reviewer comments

- ExtensionRegistry.update() now preserves original installed_at timestamp
- Add ExtensionRegistry.restore() for rollback (entry was removed)
- Clean up wrongly installed extension on ID mismatch before rollback
- Remove unused catalog_error parameter from _print_extension_info()

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(extensions): check _install_allowed for updates, preserve backup on failed rollback

- Skip automatic updates for extensions from catalogs with install_allowed=false
- Only delete backup directory on successful rollback, preserve it on failure
  for manual recovery

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(extensions): address reviewer feedback on update/rollback logic

- Hook rollback: handle empty backup_hooks by checking `is not None`
  instead of truthiness (falsy empty dict would skip hook cleanup)
- extension_info: use resolved_installed_id for catalog lookup when
  extension was found by display name (prevents wrong catalog match)
- Rollback: always remove extension dir first, then restore if backup
  exists (handles case when no original dir existed before update)
- Validate extension ID from ZIP before installing, not after
  (avoids side effects of installing wrong extension before rollback)
- Preserve enabled state during updates: re-apply disabled state and
  hook enabled flags after successful update
- Optimize _resolve_catalog_extension: pass query to catalog.search()
  instead of fetching all extensions
- update() now merges metadata with existing entry instead of replacing
  (preserves fields like registered_commands when only updating enabled)
- Add tests for ExtensionRegistry.update() and restore() methods:
  - test_update_preserves_installed_at
  - test_update_merges_with_existing
  - test_update_raises_for_missing_extension
  - test_restore_overwrites_completely
  - test_restore_can_recreate_removed_entry

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* docs(extensions): update RFC to reflect implemented status

- Change status from "Draft" to "Implemented"
- Update all Implementation Phases to show completed items
- Add new features implemented beyond original RFC:
  - Display name resolution for all commands
  - Ambiguous name handling with tables
  - Atomic update with rollback
  - Pre-install ID validation
  - Enabled state preservation
  - Registry update/restore methods
  - Catalog error fallback
  - _install_allowed flag
  - Cache invalidation
- Convert Open Questions to Resolved Questions with decisions
- Add remaining Open Questions (sandboxing, signatures) as future work
- Fix table of contents links

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(extensions): address third round of PR review comments

- Refactor extension_info to use _resolve_installed_extension() helper
  with new allow_not_found parameter instead of duplicating resolution logic
- Fix rollback hook restoration to not create empty hooks: {} in config
  when original config had no hooks section
- Fix ZIP pre-validation to handle nested extension.yml files (GitHub
  auto-generated ZIPs have structure like repo-name-branch/extension.yml)
- Replace unused installed_manifest variable with _ placeholder
- Add display name to update status messages for better UX

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(extensions): address fourth round of PR review comments

Rollback fixes:
- Preserve installed_at timestamp after successful update (was reset by
  install_from_zip calling registry.add)
- Fix rollback to only delete extension_dir if backup exists (avoids
  destroying valid installation when failure happens before modification)
- Fix rollback to remove NEW command files created by failed install
  (files that weren't in original backup are now cleaned up)
- Fix rollback to delete hooks key entirely when backup_hooks is None
  (original config had no hooks key, so restore should remove it)

Cross-command consistency fix:
- Add display name resolution to `extension add` command using
  _resolve_catalog_extension() helper (was only in `extension info`)
- Use resolved extension ID for download_extension() call, not original
  argument which may be a display name

Security fix (fail-closed):
- Malformed catalog config (empty/missing URLs) now raises ValidationError
  instead of silently falling back to built-in catalogs

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* fix(lint): address ruff linting errors and registry.update() semantics

- Remove unused import ExtensionError in extension_info
- Remove extraneous f-prefix from strings without placeholders
- Use registry.restore() instead of registry.update() for installed_at
  preservation (update() always preserves existing installed_at, ignoring
  our override)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

---------

Co-authored-by: iamaeroplane <michal.bachorik@gmail.com>
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Michal Bachorik
2026-03-13 13:23:37 +01:00
committed by GitHub
parent 82f8a13f83
commit 58ce653908
4 changed files with 1238 additions and 288 deletions

View File

@@ -1,9 +1,9 @@
# RFC: Spec Kit Extension System
**Status**: Draft
**Status**: Implemented
**Author**: Stats Perform Engineering
**Created**: 2026-01-28
**Updated**: 2026-01-28
**Updated**: 2026-03-11
---
@@ -24,8 +24,9 @@
13. [Security Considerations](#security-considerations)
14. [Migration Strategy](#migration-strategy)
15. [Implementation Phases](#implementation-phases)
16. [Open Questions](#open-questions)
17. [Appendices](#appendices)
16. [Resolved Questions](#resolved-questions)
17. [Open Questions (Remaining)](#open-questions-remaining)
18. [Appendices](#appendices)
---
@@ -1504,203 +1505,225 @@ AI agent registers both names, so old scripts work.
## Implementation Phases
### Phase 1: Core Extension System (Week 1-2)
### Phase 1: Core Extension System ✅ COMPLETED
**Goal**: Basic extension infrastructure
**Deliverables**:
- [ ] Extension manifest schema (`extension.yml`)
- [ ] Extension directory structure
- [ ] CLI commands:
- [ ] `specify extension list`
- [ ] `specify extension add` (from URL)
- [ ] `specify extension remove`
- [ ] Extension registry (`.specify/extensions/.registry`)
- [ ] Command registration (Claude only initially)
- [ ] Basic validation (manifest schema, compatibility)
- [ ] Documentation (extension development guide)
- [x] Extension manifest schema (`extension.yml`)
- [x] Extension directory structure
- [x] CLI commands:
- [x] `specify extension list`
- [x] `specify extension add` (from URL and local `--dev`)
- [x] `specify extension remove`
- [x] Extension registry (`.specify/extensions/.registry`)
- [x] Command registration (Claude and 15+ other agents)
- [x] Basic validation (manifest schema, compatibility)
- [x] Documentation (extension development guide)
**Testing**:
- [ ] Unit tests for manifest parsing
- [ ] Integration test: Install dummy extension
- [ ] Integration test: Register commands with Claude
- [x] Unit tests for manifest parsing
- [x] Integration test: Install dummy extension
- [x] Integration test: Register commands with Claude
### Phase 2: Jira Extension (Week 3)
### Phase 2: Jira Extension ✅ COMPLETED
**Goal**: First production extension
**Deliverables**:
- [ ] Create `spec-kit-jira` repository
- [ ] Port Jira functionality to extension
- [ ] Create `jira-config.yml` template
- [ ] Commands:
- [ ] `specstoissues.md`
- [ ] `discover-fields.md`
- [ ] `sync-status.md`
- [ ] Helper scripts
- [ ] Documentation (README, configuration guide, examples)
- [ ] Release v1.0.0
- [x] Create `spec-kit-jira` repository
- [x] Port Jira functionality to extension
- [x] Create `jira-config.yml` template
- [x] Commands:
- [x] `specstoissues.md`
- [x] `discover-fields.md`
- [x] `sync-status.md`
- [x] Helper scripts
- [x] Documentation (README, configuration guide, examples)
- [x] Release v3.0.0
**Testing**:
- [ ] Test on `eng-msa-ts` project
- [ ] Verify spec→Epic, phase→Story, task→Issue mapping
- [ ] Test configuration loading and validation
- [ ] Test custom field application
- [x] Test on `eng-msa-ts` project
- [x] Verify spec→Epic, phase→Story, task→Issue mapping
- [x] Test configuration loading and validation
- [x] Test custom field application
### Phase 3: Extension Catalog (Week 4)
### Phase 3: Extension Catalog ✅ COMPLETED
**Goal**: Discovery and distribution
**Deliverables**:
- [ ] Central catalog (`extensions/catalog.json` in spec-kit repo)
- [ ] Catalog fetch and parsing
- [ ] CLI commands:
- [ ] `specify extension search`
- [ ] `specify extension info`
- [ ] Catalog publishing process (GitHub Action)
- [ ] Documentation (how to publish extensions)
- [x] Central catalog (`extensions/catalog.json` in spec-kit repo)
- [x] Community catalog (`extensions/catalog.community.json`)
- [x] Catalog fetch and parsing with multi-catalog support
- [x] CLI commands:
- [x] `specify extension search`
- [x] `specify extension info`
- [x] `specify extension catalog list`
- [x] `specify extension catalog add`
- [x] `specify extension catalog remove`
- [x] Documentation (how to publish extensions)
**Testing**:
- [ ] Test catalog fetch
- [ ] Test extension search/filtering
- [ ] Test catalog caching
- [x] Test catalog fetch
- [x] Test extension search/filtering
- [x] Test catalog caching
- [x] Test multi-catalog merge with priority
### Phase 4: Advanced Features (Week 5-6)
### Phase 4: Advanced Features ✅ COMPLETED
**Goal**: Hooks, updates, multi-agent support
**Deliverables**:
- [ ] Hook system (`hooks` in extension.yml)
- [ ] Hook registration and execution
- [ ] Project extensions config (`.specify/extensions.yml`)
- [ ] CLI commands:
- [ ] `specify extension update`
- [ ] `specify extension enable/disable`
- [ ] Command registration for multiple agents (Gemini, Copilot)
- [ ] Extension update notifications
- [ ] Configuration layer resolution (project, local, env)
- [x] Hook system (`hooks` in extension.yml)
- [x] Hook registration and execution
- [x] Project extensions config (`.specify/extensions.yml`)
- [x] CLI commands:
- [x] `specify extension update` (with atomic backup/restore)
- [x] `specify extension enable/disable`
- [x] Command registration for multiple agents (15+ agents including Claude, Copilot, Gemini, Cursor, etc.)
- [x] Extension update notifications (version comparison)
- [x] Configuration layer resolution (project, local, env)
**Additional features implemented beyond original RFC**:
- [x] **Display name resolution**: All commands accept extension display names in addition to IDs
- [x] **Ambiguous name handling**: User-friendly tables when multiple extensions match a name
- [x] **Atomic update with rollback**: Full backup of extension dir, commands, hooks, and registry with automatic rollback on failure
- [x] **Pre-install ID validation**: Validates extension ID from ZIP before installing (security)
- [x] **Enabled state preservation**: Disabled extensions stay disabled after update
- [x] **Registry update/restore methods**: Clean API for enable/disable and rollback operations
- [x] **Catalog error fallback**: `extension info` falls back to local info when catalog unavailable
- [x] **`_install_allowed` flag**: Discovery-only catalogs can't be used for installation
- [x] **Cache invalidation**: Cache invalidated when `SPECKIT_CATALOG_URL` changes
**Testing**:
- [ ] Test hooks in core commands
- [ ] Test extension updates (preserve config)
- [ ] Test multi-agent registration
- [x] Test hooks in core commands
- [x] Test extension updates (preserve config)
- [x] Test multi-agent registration
- [x] Test atomic rollback on update failure
- [x] Test enabled state preservation
- [x] Test display name resolution
### Phase 5: Polish & Documentation (Week 7)
### Phase 5: Polish & Documentation ✅ COMPLETED
**Goal**: Production ready
**Deliverables**:
- [ ] Comprehensive documentation:
- [ ] User guide (installing/using extensions)
- [ ] Extension development guide
- [ ] Extension API reference
- [ ] Migration guide (core → extension)
- [ ] Error messages and validation improvements
- [ ] CLI help text updates
- [ ] Example extension template (cookiecutter)
- [ ] Blog post / announcement
- [ ] Video tutorial
- [x] Comprehensive documentation:
- [x] User guide (EXTENSION-USER-GUIDE.md)
- [x] Extension development guide (EXTENSION-DEV-GUIDE.md)
- [x] Extension API reference (EXTENSION-API-REFERENCE.md)
- [x] Error messages and validation improvements
- [x] CLI help text updates
**Testing**:
- [ ] End-to-end testing on multiple projects
- [ ] Community beta testing
- [ ] Performance testing (large projects)
- [x] End-to-end testing on multiple projects
- [x] 163 unit tests passing
---
## Open Questions
## Resolved Questions
### 1. Extension Namespace
The following questions from the original RFC have been resolved during implementation:
### 1. Extension Namespace ✅ RESOLVED
**Question**: Should extension commands use namespace prefix?
**Options**:
**Decision**: **Option C** - Both prefixed and aliases are supported. Commands use `speckit.{extension}.{command}` as canonical name, with optional aliases defined in manifest.
- A) Prefixed: `/speckit.jira.specstoissues` (explicit, avoids conflicts)
- B) Short alias: `/jira.specstoissues` (shorter, less verbose)
- C) Both: Register both names, prefer prefixed in docs
**Recommendation**: C (both), prefixed is canonical
**Implementation**: The `aliases` field in `extension.yml` allows extensions to register additional command names.
---
### 2. Config File Location
### 2. Config File Location ✅ RESOLVED
**Question**: Where should extension configs live?
**Options**:
**Decision**: **Option A** - Extension directory (`.specify/extensions/{ext-id}/{ext-id}-config.yml`). This keeps extensions self-contained and easier to manage.
- A) Extension directory: `.specify/extensions/jira/jira-config.yml` (encapsulated)
- B) Root level: `.specify/jira-config.yml` (more visible)
- C) Unified: `.specify/extensions.yml` (all extension configs in one file)
**Recommendation**: A (extension directory), cleaner separation
**Implementation**: Each extension has its own config file within its directory, with layered resolution (defaults → project → local → env vars).
---
### 3. Command File Format
### 3. Command File Format ✅ RESOLVED
**Question**: Should extensions use universal format or agent-specific?
**Options**:
**Decision**: **Option A** - Universal Markdown format. Extensions write commands once, CLI converts to agent-specific format during registration.
- A) Universal Markdown: Extensions write once, CLI converts per-agent
- B) Agent-specific: Extensions provide separate files for each agent
- C) Hybrid: Universal default, agent-specific overrides
**Recommendation**: A (universal), reduces duplication
**Implementation**: `CommandRegistrar` class handles conversion to 15+ agent formats (Claude, Copilot, Gemini, Cursor, etc.).
---
### 4. Hook Execution Model
### 4. Hook Execution Model ✅ RESOLVED
**Question**: How should hooks execute?
**Options**:
**Decision**: **Option A** - Hooks are registered in `.specify/extensions.yml` and executed by the AI agent when it sees the hook trigger. Hook state (enabled/disabled) is managed per-extension.
- A) AI agent interprets: Core commands output `EXECUTE_COMMAND: name`
- B) CLI executes: Core commands call `specify extension hook after_tasks`
- C) Agent built-in: Extension system built into AI agent (Claude SDK)
**Recommendation**: A initially (simpler), move to C long-term
**Implementation**: `HookExecutor` class manages hook registration and state in `extensions.yml`.
---
### 5. Extension Distribution
### 5. Extension Distribution ✅ RESOLVED
**Question**: How should extensions be packaged?
**Options**:
**Decision**: **Option A** - ZIP archives downloaded from GitHub releases (via catalog `download_url`). Local development uses `--dev` flag with directory path.
- A) ZIP archives: Downloaded from GitHub releases
- B) Git repos: Cloned directly (`git clone`)
- C) Python packages: Installable via `uv tool install`
**Recommendation**: A (ZIP), simpler for non-Python extensions in future
**Implementation**: `ExtensionManager.install_from_zip()` handles ZIP extraction and validation.
---
### 6. Multi-Version Support
### 6. Multi-Version Support ✅ RESOLVED
**Question**: Can multiple versions of same extension coexist?
**Decision**: **Option A** - Single version only. Updates replace the existing version with atomic rollback on failure.
**Implementation**: `extension update` performs atomic backup/restore to ensure safe updates.
---
## Open Questions (Remaining)
### 1. Sandboxing / Permissions (Future)
**Question**: Should extensions declare required permissions?
**Options**:
- A) Single version: Only one version installed at a time
- B) Multi-version: Side-by-side versions (`.specify/extensions/jira@1.0/`, `.specify/extensions/jira@2.0/`)
- C) Per-branch: Different branches use different versions
- A) No sandboxing (current): Extensions run with same privileges as AI agent
- B) Permission declarations: Extensions declare `filesystem:read`, `network:external`, etc.
- C) Opt-in sandboxing: Organizations can enable permission enforcement
**Recommendation**: A initially (simpler), consider B in future if needed
**Status**: Deferred to future version. Currently using trust-based model where users trust extension authors.
---
### 2. Package Signatures (Future)
**Question**: Should extensions be cryptographically signed?
**Options**:
- A) No signatures (current): Trust based on catalog source
- B) GPG/Sigstore signatures: Verify package integrity
- C) Catalog-level verification: Catalog maintainers verify packages
**Status**: Deferred to future version. `checksum` field is available in catalog schema but not enforced.
---