deepseek-harness/packages/client/ui-skill

@deepseek-ai/dsh-client-ui-skill

English | 中文

Skill invocation source, browser half: registers the /-trigger skill source into ctx.slash. Ordinary-session candidates come from the skill.list RPC addressed by the per-call ClientSessionContext projection's {sessionId}, with the host resolving cwd from the session header. The host serves every user-invocable skill; a modelInvocable: false entry (a disable-model-invocation skill, whose only entry point is this path) wears the user-only marker as a description prefix in the active language. Catalog-addressed continuable children resolve no skill candidates locally because the existing skill RPC requires an attached session; viewing their persisted history must not activate them. Catalogs cache per ordinary session with a single-flight fetch; the scope-birth warm hook prewarms the session's entry and connection/reset clears everything. Results filter by startsWith(query).

A menu pick or an entered /name [args] line claims the composer into an args-tolerant skill.invoke transaction (matchEnter strong-waits the catalog; an unknown name answers undefined and stays a plain prompt). Submit trims the args, keeps blank args off the wire, and folds an RPC refusal into the composer's error outcome; the host renders the skill body and injects it as a user message before starting the turn, so invocation is deterministic for every user-invocable skill. The RPC rides the plugin's root-context connection captured at registration — the source never reads services off a per-call argument. Draft chip visuals still derive from the lexicon scan; the legacy <skill>name</skill> reference codec is gone (decision 21 removal cut) and matchSpace stays unimplemented — menu and enter own the skill flows.

A failed skill.list throws from candidates, which the slash shell logs and folds into a silent menu-group drop — the menu shows only pending/ready states.

The /client export surface is the plugin body (apply/inject) only; the source object is internal to the registration effect.

Skill tool row

The browser plugin also registers a keyed skill toolview in conversation.chat.toolview. A collapsed row renders the 14-pixel skill document-and-sparkle glyph, Skill title, separator, and requested skill name with the same neutral hierarchy as the Bash row; running calls carry the transcript shimmer, failures replace the name with the first error line, and interrupted calls use the warning state. A settled row expands as one whole-row disclosure into a bounded Instructions card containing the exact durable tool output, with the standard trajectory Inspect affordance when available. The row derives its name, lifecycle, and body only from a paired call/result slice in the current runtime window, never from the current catalog, so replay remains stable when installed skills or their descriptions change.

Model Experience

User-explicit skill invocation

What the model sees

A claimed invocation never ships the /name literal. The host (skill.invoke) renders the canonical <skill_content> block — the same renderSkillContent output the skill tool returns — appends the user's trailing text after a blank line, and injects the whole as one user-role message carrying the skill-invocation source, immediately starting a turn. Loading is deterministic: the model receives the full body without being asked to call the skill tool, and the catalog (rendered by dsh-tool-skill) tells it not to re-load an inline-injected skill.

Token effect

One invocation adds the rendered skill body plus the trailing text to that turn's user message — the same cost as the model loading the skill through the tool, paid unconditionally instead of at the model's discretion. Menu browsing and the candidate fetch add zero model tokens.

KV Cache effect

Append-only: the injected message lands after the reusable history prefix. This package never edits earlier request tokens.

Known Limitations and Deferred Work

• Result-only history pages use the generic row — keyed dispatch needs the paired call in the runtime window; pagination that leaves the call outside has no tool identity. This client presentation feature does not extend the history wire contract to recover it.
• Enter waits on the catalog once — matchEnter strong-waits the session's first catalog fetch before answering, so an enter racing a cold cache resolves against the settled catalog rather than silently missing. A menu opened before the prewarm settles still shows no skill candidates for that keystroke.
• Text is the truth — the reference is plain draft text; a hand-typed identical token is the same reference. Chip visuals derive from the lexicon scan; no occurrence identity or position tracking (componentized chips are a ledger item).