- SectionLabel(): dim section headers for visual grouping in batch output
- StepFail(): preserve pre-styled ANSI labels instead of double-wrapping
- Spinner Success/Warn: skip elapsed suffix when <50ms (avoids noise in tests)
- InstallOptions.AuditVerbose field for --audit-verbose flag plumbing
- Update install integration test assertions for batch compact format
- Extend .skillshare-manifest.json tracking from copy mode to merge mode.
- Update PruneOrphanLinks to prioritize manifest records over naming patterns.
- Ensure SyncTargetMerge writes to manifest for all linked/updated skills.
- Add integration tests for uninstall + sync orphan directory cleanup.
- Maintain backward compatibility for targets without existing manifests.
Closes#42
Some AI CLIs (Cursor, Antigravity, Copilot, OpenCode) skip symlinks
when scanning skill directories, causing symlinked skills to be
silently ignored during auto-discovery.
Changes:
- `init --mode <merge|copy|symlink>` / `-m`: set default sync mode at
init time (global and project); interactive prompt when TTY detected
- `init --discover --mode`: applies mode only to newly added targets,
leaving existing targets unchanged
- Per-target mode hint after `sync` and `doctor`: suggests `copy` for
known compatibility-sensitive targets (cursor → antigravity →
copilot → opencode); suppressed when only claude/codex configured
- Neutral decision matrix in sync-modes.md for constraint-based choice
- By-target override examples in target.md, sync.md, first-sync.md
- Integration tests for mode flag, discover+mode, hint presence/absence
InstallTrackedRepo and updateTrackedRepo had no security audit,
allowing --track to bypass all supply-chain checks. This adds
auditTrackedRepo (removes repo on block for fresh installs) and
auditTrackedRepoUpdate (git reset rollback for updates), plus
input validation, testable removeAll, and hardened error handling
across audit/install/uninstall paths.
Replace simple regex-based markdown link extraction with a full
structural parser that handles inline links with titles, autolinks,
reference-style links, HTML anchors, multiline links, and correctly
skips code fences, inline code spans, and image links.
This ensures external-link and source-repository-link audit rules
match all Markdown link syntax variants while avoiding false positives
from code examples and images.
Return auditRiskScore and auditRiskLabel in update results so the
dashboard can show security posture after both tracked-repo and
regular-skill updates. auditGateTrackedRepo now returns the audit
result alongside the blocked indicator.
Regular skills (non-git, metadata-based) now support --diff flag using
hash-based file comparison, display audit results after update, and
properly enforce the audit gate by not passing Force to inner Install.
- cmd/skillshare/update_test.go: table-driven test covering all
isSecurityError keywords (security audit, post-update audit,
rolled back, rollback failed) and negative cases
- internal/server/handler_update_test.go: three tests for
auditGateTrackedRepo: rollback failure warning on HIGH findings,
rollback failure warning on scan error, and clean repo returns nil
- handler_update.go: check git.ResetHard return value; if rollback
fails, append explicit warning to blocked message instead of
silently claiming rolled back
- update.go: add "post-update audit" to isSecurityError keywords
so install.Install audit failures are correctly classified as
security blocks in batch exit code
- P1: Batch/all update now returns non-zero when repos are blocked by
security audit (fail-closed for CI pipelines)
- P2: Pass --skip-audit to regular skill update via InstallOptions
- P2: Add path containment check in content integrity to prevent
directory traversal via crafted file_hashes keys
- P3: Switch git diff parsing to NUL-delimited (-z) for safe handling
of special filenames
Pin sha256 file hashes in .skillshare-meta.json during install/update.
Audit now detects content-tampered (MEDIUM), content-missing (LOW),
and content-unexpected (LOW) findings. Backward-compatible: skills
without file_hashes in meta are silently skipped.
Supply-chain hardening Phase 2:
- Add post-pull audit gate: HIGH/CRITICAL findings auto-rollback in
non-interactive mode (fail-closed), prompt in TTY mode
- Use full 40-char commit hashes for reliable rollback (avoids short-hash ambiguity)
- Add --skip-audit flag to bypass the gate when risk is accepted
- Add --diff flag to show file-level change summary after tracked repo update
- Add GetChangedFiles() and ResetHard() git helpers
- Add audit gate in install --update git-pull path to prevent bypass
Supply-chain hardening Phase 1:
- Add source-repository-link-0 rule (HIGH) to detect [source repository](url) patterns
- Exclude source-repo labels from external-link-0 to prevent double-reporting
- Introduce risk floor: RiskLabel = max(score-label, severity-floor) so a
single HIGH finding produces "high" risk instead of "low"
- Apply severity floor to CLI summary, server API summary, and per-result risk
Inspired by #39 — batch lines, summary box, severity counts, and
single-skill risk all use a consistent color ramp:
CRITICAL → red, HIGH → orange, MEDIUM → yellow, LOW/INFO → gray.
Also hide elapsed time when < 1s for cleaner batch output.
Deduplicate bounded-concurrency worker pool that was repeated in
auditInstalled(), auditFiltered(), and handleAuditAll(). The new
audit.ParallelScan() centralises the pattern with index-aligned
results, reducing ~90 lines to a single reusable function.
Add token auto-injection for private repo support across all platforms:
- Hub search (--hub): normalize web URLs to raw-content URLs and apply
platform-specific auth headers (GitLab, GitHub, Bitbucket)
- Web UI push/pull/update: use *WithAuth variants for git operations
- Export ResolveTokenForURL and DetectPlatformForURL from install/auth
Detect `[text](https://...)` patterns in skill files as a LOW severity
finding. External links may indicate prompt injection vectors or waste
tokens. Localhost/127.0.0.1/0.0.0.0 links are excluded via the rule's
exclude regex.
Closes#38
Use bounded goroutine pool (max 8 workers) for audit.ScanSkill calls
in both auditInstalled (CLI) and handleAuditAll (API). Scan results
are collected into a pre-allocated slice, then processed sequentially
for deterministic output ordering and summary accumulation.
Add LC_ALL=C to git command environments in push, init commit, and
remote fetch to prevent locale-dependent output from breaking string
matching (e.g. "nothing to commit" in non-English locales).
Parallelize audit.ScanSkill in BuildIndex using goroutines with a
bounded semaphore (max 8 workers) to reduce index generation time
for large skill collections.
Variadic ...bool was used for backward compatibility but is non-idiomatic
Go for an internal API. Plain bool makes the function contract explicit
and simplifies the audit check from len(auditSkills) > 0 && auditSkills[0]
to a direct boolean test.
SKILL.md frontmatter may contain "version: v0.15.0" (with v prefix),
but callers already prepend "v" when formatting. Normalize at the
source by trimming the prefix in both ReadLocalSkillVersion and
FetchRemoteSkillVersion.
Scan .md files in skill directories for broken relative links during
audit. Targets that don't exist on disk produce a LOW severity finding
(pattern: dangling-link). External URLs, anchors, and fragments are
skipped.
Integrated into the existing walk (zero extra IO) and disableable via
audit-rules.yaml (- id: dangling-link / enabled: false). The init-rules
template, built-in skill reference, and website docs are updated.
Fixes multiple issues with the cross-machine sync workflow reported
in runkids/skillshare#35:
- push/pull auto-detect missing upstream and set it on first use
(push uses -u, pull uses fetch + reset + set-upstream-to)
- First pull checks local/remote skill directories before resetting;
refuses to overwrite local skills unless --force is passed
- init --remote auto-initializes git when config exists but .git does not
- HTTPS token auth (GITHUB_TOKEN, GITLAB_TOKEN, etc.) injected into
push, pull, and init remote fetch
- init --help shows init-specific flags and examples
- Remote repos with only files (README) no longer misdetected as
having skills (uses git ls-tree -d)
- Error hints suggest skillshare init --remote instead of raw git
Automatically detect GITHUB_TOKEN, GITLAB_TOKEN, BITBUCKET_TOKEN, or
SKILLSHARE_GIT_TOKEN and inject credentials via transient GIT_CONFIG_*
env vars during clone and pull. Tokens are never persisted to disk.
Key changes:
- Platform detection (GitHub/GitLab/Bitbucket) with per-platform
username conventions (x-access-token, oauth2, x-token-auth)
- GIT_CONFIG_COUNT safe append — reads existing count to avoid
overriding CI pipeline git config entries
- Token-aware error messages: "no token" shows 3 options (SSH, env
var, credential helper); "token rejected" suggests checking
permissions and expiry
- sanitizeTokens redacts all credential values (including
BITBUCKET_USERNAME) from stderr before surfacing errors
- gitPull also injects auth via getRemoteURL(), fixing the previous
limitation where update failed after HTTPS+token install
Unreadable files or unresolvable paths now cause DirChecksum to return
an error rather than silently producing an incomplete checksum.
Ref: PR #33 review (gemini-code-assist)
- SyncTargetCopy replaces non-directory managed entries with proper
skill directories; skips local non-directory entries unless --force
- CheckStatusCopy only counts managed entries that are actual
directories on disk
- diff and server diff handlers report "target entry is not a
directory" for both managed and unmanaged non-directory entries
- Add tests for non-directory edge cases in sync, diff, and target info
- P1: diff now checks target skill directory exists before comparing
checksums; deleted directories are reported as "missing" instead of
being silently treated as synced
- P2: CheckStatusCopy counts only managed entries that exist on disk,
so target info shows accurate managed count
- Add tests for deleted target dir detection and manifest/disk mismatch
- P1: diff now compares checksums to detect content drift in copy mode
- P2: diff routes on config mode instead of manifest presence
- P2: project target remove cleans up manifest file
- P3: target info uses mode-appropriate status check (copy/merge/symlink)
- Export DirChecksum for use by diff and server handlers
- Add 4 integration tests covering all fixed scenarios
Add a third sync mode (`copy`) that copies skills as real files instead
of symlinks, with a `.skillshare-manifest.json` tracking managed skills
and SHA256 checksums for incremental updates.
- Manifest-based tracking: skip unchanged, update modified, --force all
- Pruning: remove managed copies no longer in source
- Mode conversion: bidirectional merge↔copy and symlink↔copy
- Full command support: sync, status, doctor, diff, target, collect
- Server API + Web UI: targets list, drift detection, filter editing
- include/exclude filters work in both merge and copy modes
- 12 integration tests covering all acceptance criteria
- Documentation updated across 13 pages
Closes#31
- Rename "agents" target to "universal" with backward-compat alias
- Fix kimi global_path: ~/.kimi/skills → ~/.config/agents/skills
(matches official docs recommended path)
- Add replit global_name and global_path (~/.config/agents/skills)
- Prefer "universal" as canonical name in GroupedProjectTargets
- Update tests for new naming and kimi as universal group member
Enable IDE autocompletion, validation, and hover docs for config.yaml
by adding JSON Schema support via yaml-language-server directive.
- Add schemas/config.schema.json (global) and project-config.schema.json
- Prepend schema comment in Config.Save() and ProjectConfig.Save()
- Add unit and integration tests for schema comment roundtrip
- Document IDE setup in configuration.md
Change matchSkillIgnore to compare against skill path instead of name,
enabling patterns like "feature-radar" to exclude all skills under
that group directory. Precise paths like "feature-radar/helper" still
target individual skills. Backward compatible — flat skills have
path == name so existing patterns work unchanged.
The --into flag information was previously encoded only in the name
field (e.g. name: frontend/pdf), causing config-based install to fail
because validate.SkillName() rejects slashes. Add an explicit group
field to separate placement from identity, with automatic migration
of legacy slash-in-name entries.
- Add Group field, FullName(), and EffectiveParts() to SkillEntry
- Update reconcilers to write new format and migrate legacy entries
- Fix installFromGlobalConfig/installFromProjectConfig to use
EffectiveParts() for validation and FullName() for path construction
- Fix uninstall config cleanup to match via FullName()
- Add unit tests for SkillEntry helpers and reconciler group behavior
- Add 5 integration tests for group field end-to-end flows
- Document group field in declarative-manifest.md
Implement three install features with full documentation:
- .skillignore: repo-level file to hide skills from discovery (exact
match + trailing wildcard). Applied during git repo discovery before
any selection prompt.
- --exclude: CLI flag to skip specific skills during multi-skill install.
Filters before the interactive prompt so excluded skills never appear.
Warns when used with direct local-path installs (not applicable).
- License display: shows SKILL.md license frontmatter in selection
prompts and single-skill confirmation screen.
Add 7 integration tests, 2 new doc pages (declarative-manifest,
url-formats), and cross-reference updates across 6 existing doc pages.
Add detection for real-world attack vectors that were missing from the
audit engine: dynamic code evaluation (eval/exec/Function), Python shell
execution (os.system/subprocess), environment variable leaking
(process.env), prompt injection hidden in HTML comments, and escape
sequence obfuscation (hex/unicode). Each rule includes false-positive
guards (e.g. evaluate() excluded from eval rule, NODE_ENV excluded from
env-access). Also backfills insecure-http and shell-chain rules that
were missing from the website docs table.
Firebender is a coding agent for JetBrains IDEs with native skills support.
Paths: ~/.firebender/skills
(global), .firebender/skills (project). Update target count to 49+.