- Resolve skills root via Path.resolve() so LLM prompts always
reference absolute paths regardless of sandbox cwd
- Use resolved path in skill metadata for reliable cat/head commands
- Add DRY cross-reference comment for frontmatter parser
- Remove dead skills_root_abs field from JSON output (no consumer)
- Remove unnecessary os import and fake resolve/abspath branch
- Rewrite build_skills_prompt() with structured numbered rules and
markdown formatting for better LLM comprehension
- Sanitize example_path with _SAFE_PATH_RE before embedding in system
prompt to prevent prompt injection via crafted skill paths
- Add docstring to _parse_frontmatter_description()
- Remove debug print(top_dirs) from install_skill_from_zip()
- Remove stale commented-out SANDBOX_SKILLS_ROOT line
_apply_sandbox_tools now checks the booted session's capabilities
before registering browser tools (BrowserExecTool, BrowserBatchExecTool,
RunBrowserSkillTool).
- If no session exists yet (first request), all tools are registered
conservatively to avoid breaking the initial interaction
- If a session exists without browser capability, browser tools are
omitted, preventing CapabilityNotSupportedError from Bay
- Skill lifecycle tools remain unconditionally registered
Add capabilities property to ComputerBooter base class (returns None)
and ShipyardNeoBooter (returns immutable tuple from sandbox).
- Extract DEFAULT_PROFILE class constant to replace scattered magic string
- Use tuple[str, ...] for immutability (no defensive copy needed)
- Add _resolve_profile() for smart profile selection:
- honour user-specified profile
- query Bay API, prefer browser-capable profiles
- re-raise auth errors (401/403), fallback on transient failures
- Conditionally create NeoBrowserComponent only when profile has browser
- Log resolved profile and capabilities at boot
Add BayContainerManager to manage Bay container lifecycle via Docker
Engine API, similar to how BoxliteBooter manages Ship containers.
When ShipyardNeoBooter endpoint is empty or set to '__auto__', Bay is
automatically pulled, started, health-checked, and credentials are
read from the container.
- New bay_manager.py: ensure_running, wait_healthy, read_credentials
- Integrate auto-start into ShipyardNeoBooter boot/shutdown
- Reuse Bay container across sessions (unless-stopped policy)
- Friendly error messages for Docker and credential failures
When saving config with shipyard_neo sandbox, _validate_neo_connectivity()
performs an async /health check against the Bay endpoint. If Bay is
unreachable, a ⚠️ warning is appended to the success snackbar message.
Config still saves successfully — the warning is informational only.
- Add _discover_bay_credentials() auto-discovery in _get_neo_client_config()
- Catch ValueError separately in _with_neo_client(), log at DEBUG instead of
ERROR with full traceback — prevents log spam when visiting Skills page
without Bay configured
separate sandbox skill syncing into distinct apply and scan steps
while keeping the legacy combined command for compatibility
improve observability by adding phase-based logs and richer shell
error details that include exit code, stderr, and stdout tail
reuse a shared python-exec command builder to reduce duplication
and keep command generation consistent
extract shared promote/sync orchestration into `NeoSkillSyncManager` so
computer tools and dashboard routes use the same rollback and error logic
add a reusable neo tool base runner to remove duplicated admin checks and
try/catch handling across skill-related tools, keeping responses consistent
factor sync result serialization into a single helper and reuse it where
stable release sync output is returned
extract a shared `_with_neo_client` wrapper to handle neo client
setup, teardown, and error responses in one place.
reduce duplicated try/except and `BayClient` context boilerplate across
neo skills endpoints while preserving existing request validation and
response payloads.
set the base booter browser property to return None instead of
raising NotImplementedError so callers can handle missing browser
support through capability checks
When shipyard_neo_access_token is not configured, _discover_bay_credentials()
searches for Bay's credentials.json in:
1. BAY_DATA_DIR env var
2. Mono-repo relative path ../pkgs/bay/
3. Current working directory
Enables zero-config dev mode when Bay runs locally alongside AstrBot.
Add _log_computer_config_changes() to detect and log modifications to
computer_use_runtime and sandbox.* keys when saving config via Dashboard.
Sensitive fields (tokens/secrets) are masked in log output.
* fix: handle list format content from OpenAI-compatible APIs
Some LLM providers (e.g., GLM-4.5V via SiliconFlow) return content as
list[dict] format like [{'type': 'text', 'text': '...'}] instead of
plain string. This causes the raw list representation to be displayed
to users.
Changes:
- Add _normalize_content() helper to extract text from various content formats
- Use json.loads instead of ast.literal_eval for safer parsing
- Add size limit check (8KB) before attempting JSON parsing
- Only convert lists that match OpenAI content-part schema (has 'type': 'text')
to avoid collapsing legitimate list-literal replies like ['foo', 'bar']
- Add strip parameter to preserve whitespace in streaming chunks
- Clean up orphan </think> tags that may leak from some models
Fixes#5124
* fix: improve content normalization safety
- Try json.loads first, fallback to ast.literal_eval for single-quoted
Python literals to avoid corrupting apostrophes (e.g., "don't")
- Coerce text values to str to handle null or non-string text fields
* feat: implement fallback provider support for chat models and update configuration
* feat: enhance provider selection display with count and chips for selected providers
* feat: update fallback chat providers to use provider settings and add warning for non-list fallback models
zenfun
RC-CHN
Ruochen Pan
Soulter
Soulter
Raven95676
NayukiMeko
Waterwzy
Li-shi-ling
エイカク
stevessr
xzj0898
evpeople