* feat(pi): add native RPC transport and compatibility contracts * feat(pi): implement native harness and event normalization * feat(pi): resume native and legacy ACP sessions * feat(pi): support native steering and reliable cancellation * feat(pi): expose native models commands and skills * feat(pi): bridge extension dialogs to desktop and mobile input * feat(pi): preserve Zeron MCP delegation in native sessions * feat(pi): activate native RPC driver and CLI discovery * refactor(pi): retire ACP adapter and verify native lifecycle parity * fix(pi): preserve switched sessions empty resumes and composer drafts * docs(pi): translate RPC migration research into English * style(ios): trim whitespace in updated question bindings * test(acp): wait for scenario progress before steering * test(pi): compare canonical session paths on Windows * fix(pi): deliver steering bursts in the same model step * fix(ios): preserve generated Swift binding whitespace * fix(pi): scrub nested-agent env and keep dropped-resolver semantics Pi's native launch skipped owned::configure, so CLAUDECODE and the other nested-agent variables the ACP launch removed leaked into pi. The engine's new input-cancellation task also turned a dropped resolver into an empty answer; keep it an error for harnesses (opencode rejects on error). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * ci: run the native Pi RPC driver tests pi_rpc needs the native-fixture feature and pi_resume is an engine test; neither ran in any workflow. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(pi): never overwrite a configured Pi steering mode set_steering_mode persists to Pi's global settings.json. Selecting `all` on every session start silently rewrote the user's standalone Pi preference and undid any `/steering` choice on the next run. Select `all` only while neither the global nor the project settings configure a mode. PiHarness::with_agent_dir lets tests (and callers) pin PI_CODING_AGENT_DIR for both the child and Zeron's own reads, so fixture tests never consult the developer's real Pi settings. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(ios): don't trap on duplicate question ids when prefilling answers Dictionary(uniqueKeysWithValues:) crashes on a repeated key; question ids come from agents and synced docs, so keep the first instead. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * style(pi): rustfmt the steering-mode guard and its tests Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(pi): start fresh with a notice when a session can't be restored The native driver hard-failed a resume whose Pi session file was missing. The engine injects the chat's stored session id on every dispatch, so one unwritten or deleted session file left the chat failing every message for good. The ACP adapter and the Codex driver start a new session and announce the lost context instead; do the same. Proven-empty sessions still keep their UUID via --session-id. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(pi): never fail a turn over an attachment Pi can't inline Composer uploads include SVG, HEIC, BMP, TIFF and AVIF. The native driver errored the whole run on any of those (or an unreadable path), while the Claude driver skips them and the retired adapter ignored attachments: the path refs already ride the prompt text. Load images best-effort instead. Also give native startup the 120s budget the ACP handshake gave Pi, since extensions can start cold. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * style(pi): rustfmt the lost-session live test Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Wing <contact@winglee.io> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
5.7 KiB
Commands and skill completion across harnesses
Workspace commands
Every harness gets Zeron actions in the / picker. Selecting one runs the action
immediately; it does not create a model turn. Actions that require a conversation
are shown only in an existing chat.
| Command | Action |
|---|---|
/model |
Choose agent, model, and reasoning |
/new |
Start a new conversation |
/resume |
Search and open conversations |
/settings |
Open Zeron settings |
/diff |
Open a changes tab |
/files |
Open project files |
/terminal |
Open a terminal tab |
/rename |
Open the conversation rename dialog |
/stop |
Interrupt the current run |
Provider commands remain available alongside these actions. When a provider owns
the same name, its command keeps the name and the Zeron action gets a zeron:
prefix, for example /zeron:model. A provider command and a workspace action are
distinct entries, even when they have the same purpose. / completion works at
any word boundary in the draft, including later lines, just like @ and $.
Code, URLs and paths remain literal.
Selecting a workspace action consumes only its trigger, preserving surrounding
text, attachments and queued edits. Removing the trigger is undoable. Provider
commands and skills insert references into the draft and are delivered on Send;
provider command execution still follows its protocol's leading-command rules.
Typing an unselected inline /word does not execute a local action.
| Harness | Additional provider commands |
|---|---|
| Codex | Native /compact and /review; eleven total entries in an existing chat before skills |
| Claude Code | Project-scoped commands from its initialize catalog |
| OpenCode | Project-scoped server command catalog |
| Pi | Native get_commands, including extension commands and skill:* invocations |
| Devin, Grok, Hermes, Antigravity | Advertised ACP commands, including session command updates |
| Cursor | Workspace actions; its SDK adapter has no native command catalog |
This does not imply that every command in a provider's terminal UI can execute through its SDK. Provider-only operations without a corresponding protocol or Zeron action are not advertised as executable commands.
Skill preferences
Settings → Shortcuts → Composer completion has two independent controls for each installed, enabled agent on this device. Agents that are disabled or not installed do not appear in this section.
- $ for skills enables the skill picker after
$. - Separate / commands removes skills from the
/picker.
Codex defaults to both enabled. Other harnesses default to both disabled, keeping
skills available through /. To use Claude Code or OpenCode with $ for skills
and / for commands, enable both controls for that harness. Enabling $ alone
makes skills available in both pickers. The preferences are saved per harness;
existing “Show skills in / menu” preferences are preserved until overridden.
These controls change completion, not arbitrary typed text. Selecting a skill stores its identity in the composer. Literal dollar signs, code, and unselected text retain their meaning.
Discovery and delivery
Discovery runs on the selected host in the project directory. Codex uses its
native skill catalog. Other adapters discover standard project and user skill
directories, including shared .agents/skills directories. Claude checks native
command availability; OpenCode adds skills from its server command catalog.
ACP adapters bind discovered skills to advertised commands, and Pi also exposes
advertised /skill:name entries without a local file.
Shared .agents/skills files use explicit file delivery in Claude and generic ACP
catalogs: a matching built-in command name does not establish skill identity.
Pi's skill: namespace and OpenCode's skill source metadata identify native skills
explicitly and remain eligible for native delivery.
At delivery, Codex receives typed skill inputs. A leading skill selection with a native command for the current harness becomes that command, preserving its arguments. Other selections include an explicit skill reference in the prompt. Native command metadata is scoped to its harness, so switching agents does not invoke another agent’s same-named command. Only catalog entries identified as skills are removed from the slash picker when separation is enabled.
File discovery covers the standard directories implemented by each adapter; custom plugin paths and provider configuration are exposed where the native catalog provides them. It is not a complete parser of every provider’s config. Unreadable files, directories, and directory entries are skipped with diagnostics so an unrelated skill root cannot suppress valid results.
Hosts advertise composer-references-v1 for canonical reference delivery.
Selecting a provider command for an older host inserts literal slash text.
Sending an existing file, command, or skill chip to an unsupported host asks the
user to update that device and preserves the draft and attachments; the same
check applies when committing an edited queued message. Literal code examples
do not require this capability.
Provider references: