Change skill mounting behaviour

[ScriptSmith]

No description provided.

[greptile-apps]

Greptile Summary

This PR changes how skill bundles are mounted in the sandbox: stored skills now mount at /skills/<name>-<version> (mirroring OpenAI's naming scheme) instead of /skills/<skill_id>, and inline skills move from /skills/skill_inline_<hash> to /skills/<name> directly. Inline skill names are now required to be lowercase slugs, which also serves as a path-traversal guard since the name is written to the sandbox verbatim.

• Replaces the hash-based synthetic-id mount path for inline skills with the plain name slug, ensuring the same stable path across foreground/background dispatch lanes without relying on content-based hashing.
• Adds validate_skill_name enforcement for inline skill names at both request time and container creation time, with a new MountPathConflict error to prevent two skills from silently clobbering each other's files at mount time.
• Updates documentation, OpenAPI schemas, and test coverage to reflect the new <name>-<version> / <name> layout.

Confidence Score: 5/5

Safe to merge; the path-traversal guard is correctly implemented and all new error paths are properly wired up.

The logic changes are well-contained: slug validation is enforced at both container creation and request resolution time, the new ensure_unique_mount_paths check prevents silent file clobbering, and the new error variants map cleanly to HTTP 400 responses. Tests cover the security-critical path-traversal case and the collision detection. The only gap — not detecting duplicate inline names at container creation — degrades UX for a specific misconfigured container but never causes silent data loss or a security bypass.

src/routes/api/containers.rs — the per-entry validation loop could be extended to detect duplicate inline skill names before persisting.

Important Files Changed

Filename	Overview
src/services/responses_pipeline.rs	Core skill resolution logic: changes stored mount path to <name>-<version>, inline path to <name>, adds InvalidInlineName / MountPathConflict error variants, ensure_unique_mount_paths helper, and path-traversal guard via validate_skill_name. Well-tested, but inline-duplicate conflicts in containers are deferred to request time rather than caught at creation.
src/routes/api/containers.rs	Adds slug validation for inline skill names at container creation time, matching what the resolver enforces at request time. Cross-skill conflict detection (two inline skills with the same name) is not checked here.
src/routes/api/chat.rs	Adds InvalidInlineName and MountPathConflict to the error-code mapping; both map to sensible HTTP 400 responses with descriptive error codes.
src/runtimes/mod.rs	Doc comment on SkillMount.mount_path updated to reflect new <name>-<version> format; no logic changes.
src/api_types/responses.rs	Doc strings updated to reflect new mount path conventions; no logic changes.
src/services/shell_tool.rs	Comment on mounted_skill_ids updated to describe new path format; no logic changes.
agent_instructions/containers.md	Documentation updated to reflect new skill reference resolution, mount paths, and inline skill slug requirements.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[RequestSkill list] --> B{Skill type?}
    B -->|skill_reference| C[resolve_version_for_reference\nDB lookup by id / name / version]
    B -->|inline| D[resolve_inline_skill\nvalidate slug name\ndecode base64]
    C --> E[mount_path = /skills/name-version_seq]
    D --> F[mount_path = /skills/name]
    E --> G[ensure_unique_mount_paths\nquadratic scan over resolved list]
    F --> G
    G -->|conflict| H[SkillResolutionError::MountPathConflict\nHTTP 400 skill_mount_conflict]
    G -->|ok| I[build_preamble per skill\nprepend to instructions]
    I --> J[Vec SkillMount to ShellExecutor\nfiles written to sandbox at mount_path]

Loading

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
src/routes/api/containers.rs:593-597
**Missing duplicate-name check for inline skills at creation time**

The comment at `validate_skill_entry` says it mirrors the resolver's checks "so a misshaped inline skill never makes it onto a stored row," and this PR adds the slug-validity check to match. However, a container created with two inline skills that share the same `name` (both pass per-entry validation) will have `skill_mount_conflict` returned on every subsequent request that uses the container, because `ensure_unique_mount_paths` is only called at request resolution time — not here.

Unlike stored-skill conflicts (which can't be predicted at creation because the resolved version is unknown), inline skill mount paths are entirely determined by `name`, so the clash is detectable right now. A simple duplicate-name scan over the inline entries in `body.skills` before persisting would catch this class of always-broken containers at creation time, consistent with the stated motivation for the existing slug check.

_{Reviews (3): Last reviewed commit: "Review fixes" | Re-trigger Greptile}

[ScriptSmith]

@greptile-apps