Files
skillshare/website/docs/reference/commands/hooks.md
T
ed3c825c49 feat(hooks): record Codex hook trust on sync with --trust (#829)
* feat(hooks): record Codex hook trust on sync with --trust

Codex runs a user-level hook only after its trusted_hash is stored in that
home's config.toml, keyed by the hook's position in hooks.json. Only the
TUI's /hooks review writes it, so every hook Skillshare adds or moves starts
untrusted and codex exec skips it without a word.

`hooks sync --trust` (global mode, opt-in) asks each Codex home's
`codex app-server` for the current hash of the hooks Skillshare owns and
writes trusted_hash through config/batchWrite, the same calls /hooks makes,
so the hash is always the one Codex computes. It changes nothing else in
config.toml: disabled hooks stay disabled and are reported. Trust it wrote
is remembered in the hooks state directory, so when a hook moves or is
removed the next --trust sync clears the old position, unless that trust
has changed since. Without codex on PATH the hook files are still written
and the command fails saying trust was not recorded.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* fix(hooks): scope Codex trust to config owner

Scope trust selection and saved hashes to the active config owner so a global config cannot trust or reconcile another config’s hooks.

Recheck hooks.json after hooks/list to avoid trusting a handler from a stale position, and only report trusted or cleared actions after config/batchWrite succeeds. Keep v1 ownerless hashes cleanup-only so upgraded installs clear stale trust after hook removal without trusting foreign hooks.

* fix(hooks): retain parked Codex trust receipts

Locate hooks using all records for their canonical file before selecting the active config owner, so identical hooks remain identifiable across shared and symlinked homes. Keep cleanup receipts for parked registrations whose native group still matches, including when duplicate counts make locate ambiguous, while clearing receipts for removed hooks.

Add regressions for identical hooks across config owners, symlink aliases, parked-home restoration, ambiguous duplicates, and stale receipt cleanup.

* fix(hooks): coalesce symlinked Codex trust homes

Canonicalize current-owner and legacy trust receipts so symlink spellings of one hooks.json are processed once. Remove the redundant parked-position branch because the hash-matching fallback below covers it. Add regression coverage for current and legacy receipts.

* fix(hooks): coalesce removed Codex hook homes

Resolve existing home directories when hook files have been removed so trust receipts for symlink aliases reconcile once. Correct canonical key expectations and exercise coalescing without unsupported ledger migration.

* fix(hooks): read reconciled trust fixture into fresh state

JSON decoding retains existing map entries, so reusing the seeded receipt map makes the alias appear to survive reconciliation. Decode the persisted state into a new value to verify its actual contents.

* feat(hooks): confirm Codex trust before recording it and drop the receipts file

`hooks sync --trust` recorded trust for every owned hook without showing
what it was trusting, and tracked its own writes in codex-trust.json so
it could clear them later. Codex's review exists so that a person sees a
command before it runs, and the receipts file was the source of most of
the edge cases found in review.

Trust is now two steps. A preview asks Codex which owned hooks are new
or changed and returns them with their commands and a revision; the
write takes that revision and is refused when the hooks changed since.
The CLI lists the hooks and asks once, `--yes` skips the question, and
without a terminal nothing is trusted unless `--yes` is given. The
dashboard gets the same two steps as API routes.

Nothing is cleared any more: a leftover entry only matches a hook with
the same event, matcher and handler at that position, which is the hook
the user already approved. That removes codex-trust.json.

The app-server is started without CODEX_HOME for the default home.
Codex resolves the symlinks of CODEX_HOME but not of ~/.codex, so a
symlinked default home got a trust key the user's own Codex never looks
up.

A home whose write fails no longer stops the other homes, and the error
names which were recorded. A preview releases the hooks operation lock
before it waits on Codex, so a slow Codex does not fail other hooks
operations.

Refs #828

* feat(ui): trust Codex hooks from the Hooks page

The dashboard could sync Codex hooks but left them untrusted, so Codex
skipped them until the user ran the CLI or approved each one in Codex.

The Sync box now offers "Trust Codex hooks" in the global scope when a
hook is bound to a Codex target. The dialog shows each hook that needs
trust with its full command and records trust for exactly that list; a
change since the list was shown is refused. The button is hidden while
changes are pending, because trust covers the hooks on disk and the next
sync would invalidate it.

Refs #828

* test(ui): cover the Codex trust dialog's refused, empty and disabled states

Only the happy path was tested, so a regression in the states where the
dialog must not offer a write would have passed. The docs sentence on
several Codex homes now says what is isolated: a failed write, not a
home that cannot be listed.

Refs #828

* refactor(hooks): split the trust prompt and output rendering out of the handlers

hooks_handlers.go had grown past 500 lines and mixed orchestration with
the confirmation prompt and every print function, which the CLI layering
rule asks to keep apart. The code is moved unchanged: the prompt goes to
hooks_prompt.go and the printing to hooks_render.go.

* fix(hooks): leave out a Codex home that cannot be listed instead of stopping

With several Codex homes, one home Codex could not list aborted the whole
trust run, so a broken account home kept every other home untrusted.

A home that fails to list is now reported and left out, and the other
homes are previewed and trusted as usual. The CLI warns before the
question and exits with an error that names the home; the dashboard
dialog shows the same note. When no home can be listed the preview still
fails.

Codex answers a home it cannot load, such as one with an invalid
config.toml, with a successful reply that has no hooks and the reason in
`errors`. That was read as a home without hooks and passed silently. It
is now a failed listing that carries Codex's own message.

Refs #828

* fix(hooks): check the reviewed Codex hooks per home when recording trust

The revision was one digest over every home's hooks. When a home was
listed in the preview but failed to list again at the write, its hooks
dropped out of the digest, so the write was refused as stale: the homes
the user had just reviewed stayed untrusted and the real listing error
was hidden behind "changed since they were listed".

The revision now has one part per home. A home is refused only when its
own hooks changed; a home that cannot be listed any more is reported and
the reviewed homes are still trusted.

Refs #828

* chore: release as 0.30.10

Keep the next release on the patch version the maintainer asked for,
instead of the 0.31.0 that Release Please derives from the feat commit.

Release-As: 0.30.10

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Willie <egg358358@gmail.com>
2026-11-17 15:54:20 +08:00

29 KiB
Raw Blame History

sidebar_position
sidebar_position
4

hooks

Manage named hooks and synchronize each receiving Agent's or Git's native configuration. Use the dashboard's Hooks page to add, edit, import, enable or disable entries, then preview before syncing. Hooks also appear in destination Targets and Projects, Sync, and Settings → Backups. The Hooks page works only when the browser opens the dashboard by localhost, an IP address or a trusted origin; see Reverse Proxy.

Use a hook row's menu to view the native configuration or code for each target, including script files and destination paths. This read-only preview shows that hook's contribution; shared files keep their other settings. Disabled hooks can be inspected but publish nothing.

Commands

skillshare hooks
skillshare hooks list --json
skillshare hooks add check --file ./check.yaml
skillshare hooks edit check --file ./updated-check.yaml --sync
skillshare hooks import --from claude --json
skillshare hooks import imported --from claude --file ./settings.json --dry-run
skillshare hooks disable check --sync
skillshare hooks enable check --sync
skillshare hooks sync --dry-run --json
skillshare hooks sync
skillshare hooks sync check --replace --dry-run
skillshare hooks sync --trust -g
skillshare hooks sync --trust --yes -g
skillshare hooks sync --trust --yes --json -g
skillshare sync hooks --dry-run --json
skillshare sync hooks
skillshare hooks remove check --sync
skillshare hooks remove check --keep-files
skillshare hooks restore BACKUP_ID --dry-run
skillshare hooks restore BACKUP_ID

Without a subcommand, hooks lists entries. add and edit read an Entry JSON or YAML document from --file. Import without a name lists candidates, one per Agent event or file; choose one name to save. Saving an import takes over the registrations it read in place: the next sync adopts them without --replace, and events you did not import stay unmanaged. Import only reads configuration or code and never runs it.

The dashboard's Hooks in targets dialog opens with no hook selected; tick the ones to import. It can also edit or remove a hook in its Agent's settings file without importing it. This works for Agents that keep hooks as events in one settings file (Claude, Codex, Gemini, Cursor, Qwen, and Droid's hooks.json); hooks that are a whole file or a named block must be imported first. The dialog shows the file diff before anything is written and backs the file up; the hook stays unmanaged.

Add, edit, import, enable, disable and remove save source only unless --sync is supplied. Sync and restore always preview again before applying. sync hooks is the resource-only alias; sync --all includes hooks and checks native conflicts before other resources change. Its JSON output has a hooks result alongside the other resource results. These are separate operations, not one transaction.

Option Meaning
--file PATH Entry JSON/YAML for add/edit; native configuration or code for import
--from AGENT Native Agent format or existing Agent/account configuration to import
--sync Save and synchronize the mutation
--keep-files With remove: stop managing the hook and leave its native entries as they are. Not with --sync. See below
--replace Explicitly replace an existing source entry or conflicting native output for that entry
--trust With hooks sync in global mode: also record Codex trust for the hooks Skillshare owns. See Codex trust
--yes, -y With --trust: skip the question and trust the listed hooks. See Codex trust
--dry-run, -n Preview without saving or writing native configuration
--json Structured output
--revision ID Require the matching preview revision
--global, -g Global Skillshare configuration
--project, -p Project-local Skillshare configuration

Use hooks list --json to see backup IDs and exact destination paths. A changed configuration invalidates the preview; refresh it before saving or syncing.

Stop managing a hook

skillshare hooks remove check --keep-files

This removes check from the source and forgets which native registrations and files Skillshare wrote for it. No Agent file changes. From then on those entries are yours: sync neither removes nor updates them, and hooks import offers them again. --keep-files cannot be combined with --sync. The dashboard's remove dialog offers it as Stop managing, on the Hooks page and in a project's Hooks tab. Remove from source only is different: the next sync deletes the hook's entries from the Agent files.

--keep-files is unavailable for Git bindings; see Git hooks.

Only the scope you remove it from changes. Stopping a global hook leaves a project's hook of the same name managed, and the other way round. To manage the entries again, import them.

Source fields

Declarations live in hooks.entries in the selected Skillshare config. A name identifies one entry; its bindings select receiving Agents and contain their native definitions. An empty bindings map keeps the entry in Skillshare only.

hooks:
  entries:
    check:
      description: Run the project's check after Claude finishes
      enabled: true
      bindings:
        claude:
          events:
            Stop:
              - hooks:
                  - type: command
                    command: "make check"
                    timeout: 120

The file passed to hooks add check --file check.yaml contains only the Entry: description, enabled and bindings, without the hooks.entries wrapper.

Skillshare writes the hooks section in indented block style, and each save unfolds entries an earlier edit left on one line. The config.yaml button on the Hooks page opens Settings → Files at the hooks: section. There, clicking a hooks key explains it in the right panel, and Beautify unfolds nested one-line sections while short lists such as targets: [claude, codex] stay on one line.

Field Meaning
description Optional description
enabled Defaults to true; false retains the source and removes unchanged owned outputs on the next sync
bindings Map of native Agent IDs to bindings
bindings.AGENT.events Native event map for command/configuration Agents
bindings.AGENT.code Supplied native extension/plugin source for Pi, OMP, Amp or OpenCode
bindings.AGENT.codeFile Instead of code: a path, relative to the config's folder, whose contents are used as code
bindings.git.commands Named Git config hooks: events, command, optional parallel
bindings.AGENT.files Optional UTF-8 script files for command bindings, keyed by relative filename; a value is the contents or {from: path}, read like codeFile

Agent IDs are claude, codex, gemini, copilot, cursor, droid, qwen, antigravity, pi, omp, amp, opencode and git. factory is accepted as an alias for droid, and antigravity-cli and agy for antigravity. A binding may also name an account target declared under targets in global mode. Keep event names, matchers, handler types, commands, timeout units and payloads in each Agent's native format. Skillshare does not translate one Agent's runtime behavior into another's. Event names are checked against each command Agent's documented events: an unknown name, such as a misspelled Stopp, is a warning in previews and in the plan's warnings, never a sync blocker, because Agents add events over time. Pi, OMP, Amp and OpenCode code is not checked.

Pi, OMP, Amp and OpenCode use their own extension/plugin APIs. Supply code matching the installed Agent version, including its imports. Skillshare writes it to a dedicated skillshare-NAME.ts file (OpenCode v1), or an OpenCode v2 directory package, without generating a universal hook runtime.

Longer code and scripts can live in their own files, which a tsconfig, shellcheck or an editor can check. Use codeFile instead of code, and {from: path} instead of a files value; inline values keep working. Both are read every time the config is read, so edit the file and run hooks sync; a change made after a preview fails the apply as stale. A path resolves against the folder of the config that holds the entry (.skillshare/ in a project, the global config's folder for hooks.projects; with hooks add|edit --file, the selected config's folder, not the entry file's), must stay inside it, and must not pass through a symlink. A missing file stops the config from loading, like any other invalid entry. Saving an entry keeps the references. Editing the code or a script in the dashboard saves it inline. JSON (hooks list --json, the Web API) carries the contents in code and files, with the references in codeFile and filesFrom; an entry sent back with both must send the file's current contents. An entry JSON file may also write a files value as {"from": path}.

hooks:
  entries:
    lint:
      bindings:
        pi:
          codeFile: hooks/lint.pi.ts
        cursor:
          events:
            afterFileEdit:
              - command: .cursor/hooks/skillshare/lint/lint.sh
          files:
            lint.sh: { from: hooks/lint.sh }

Oh My Pi code bindings

Use bindings.omp.code with the extension API of your installed OMP version. Skillshare copies the source verbatim; it does not translate shell events or execute the extension to validate it. OMP has no project trust prompt, so a synced project extension can execute at the next OMP startup.

OMP's hooks/pre and hooks/post factories remain additional unmanaged sources. They are listed but not imported into extensions, which would risk loading the same factory twice. Importing an existing standalone extension also leaves its original file in place; review the import warning before syncing another copy.

Pi and OMP both honor PI_CODING_AGENT_DIR. If bindings resolve to the same output file, Skillshare blocks the plan before writing. Use separate explicit account directories when managing both. Script files are stored under the Agent config directory's hooks/skillshare/NAME/; commands retain the native macros or explicit paths you provide. Inspect the exact paths in the preview.

Native destinations

Agent Global default Project default Native shape
Git $XDG_CONFIG_HOME/git/skillshare/hooks.gitconfig (default ~/.config/git/…) <git-common-dir>/skillshare/hooks.gitconfig Git 2.54+ hook.<name> sections
Claude Code ~/.claude/settings.json .claude/settings.json hooks event map with matcher groups
Codex ~/.codex/hooks.json .codex/hooks.json Wrapped hooks event map
Gemini CLI ~/.gemini/settings.json .gemini/settings.json hooks event map
Copilot CLI ~/.copilot/hooks/skillshare-NAME.json .github/hooks/skillshare-NAME.json Version 1, hooks event map
Cursor ~/.cursor/hooks.json .cursor/hooks.json Version 1, native lowerCamelCase events
Factory Droid ~/.factory/hooks.json .factory/hooks.json Unwrapped event map
Qwen Code ~/.qwen/settings.json .qwen/settings.json hooks event map
Antigravity ~/.gemini/config/hooks.json .agents/hooks.json Named hook blocks, one per hook
Pi ~/.pi/agent/extensions/skillshare-NAME.ts .pi/extensions/skillshare-NAME.ts Native extension code
Oh My Pi ~/.omp/agent/extensions/skillshare-NAME.ts .omp/extensions/skillshare-NAME.ts Native OMP extension code
Amp ~/.config/amp/plugins/skillshare-NAME.ts .amp/plugins/skillshare-NAME.ts Native plugin code
OpenCode ~/.config/opencode/plugins/skillshare-NAME[.ts] .opencode/plugins/skillshare-NAME[.ts] v1 file; v2 directory package (see below)

Native config-directory environment overrides apply in global scope, unless they point at an account's config_dir. Project operations write inside that project and never fall back to a global path. Other native sources, such as Codex inline TOML declarations, remain separate. Antigravity and its CLI (agy) read the same hooks.json; each hook is one block named after it, so importing a block keeps its name. Hooks in the CLI's ~/.gemini/antigravity-cli/settings.json stay separate. Copilot loads project hooks from .github/hooks only in a trusted folder. Sync refuses to create a Droid standalone file while active inline hooks exist. Import them first, review and remove the original inline hooks, then sync; this avoids silently changing which native source Droid loads.

Another account of an Agent

A target declared with agent and config_dir can receive hooks under its own name. Claude, Codex, Pi and OMP accounts use their Agent's native binding format. The git key always means Git config hooks, so an account named git receives no hooks, and the plan warns once an entry uses the git binding; give it another name:

targets:
  codex-2:
    agent: codex
    config_dir: ~/.codex-2
    skills: {enabled: false}
hooks:
  entries:
    check:
      bindings:
        codex-2:
          events:
            Stop:
              - hooks:
                  - type: command
                    command: "echo checked"
Account's Agent Native destination
claude <config_dir>/settings.json
codex <config_dir>/hooks.json
pi <config_dir>/extensions/skillshare-NAME.ts
omp <config_dir>/extensions/skillshare-NAME.ts

Command scripts go to <config_dir>/hooks/skillshare/NAME/. Accounts are global only: project configs and hooks.projects bind the Agent itself, because every account reads the same project files. A missing config_dir is skipped with a warning; sync never creates an account home.

If CODEX_HOME, CLAUDE_CONFIG_DIR or PI_CODING_AGENT_DIR points at a declared account's directory, the plain Agent binding uses its built-in default home. The plan warns and each account still uses its own config_dir. Other native directory overrides keep their existing behavior. Directories that resolve to the same location, including symlinks, cannot receive two binding keys in one sync.

Owned outputs whose target was removed, whose home moved, or whose account home is missing are left in place with their ownership records and a warning. Restore the target or directory to resume managing them. To release one entry's parked records without changing those files, preview and run skillshare hooks sync NAME --replace -g. Removing a target warns if hook bindings still name it; remove those bindings too before the next hooks sync.

Import from an account with skillshare hooks import --from codex-2 -g --json. The dashboard labels it codex-2 (Codex) and uses Codex's events and editor. Each Codex home needs its own review and trust of changed hooks in /hooks; sync never changes trust or executes a hook unless you pass --trust.

Codex trust

Codex runs a user-level hook only after it is trusted, and keeps that trust in the same home's config.toml as hooks.state."<hooks.json>:<event>:<group>:<handler>".trusted_hash. The key is the hook's position, so a hook Skillshare adds or moves starts untrusted, and codex exec skips it.

skillshare hooks sync --trust -g
skillshare hooks sync --trust --yes -g
skillshare hooks sync --trust --yes --json -g

--trust first writes the hook files as a normal sync does. Then it asks Codex which of the hooks Skillshare owns need trust, in every Codex home it manages, account homes included. It gets each hook's hash from codex app-server, the same calls the /hooks review uses, so codex must be on PATH. Skillshare never computes the hash.

Hooks that are already trusted are reported as unchanged and need no answer. If a hook is new, or its command changed since it was trusted, the command lists those hooks with their commands and asks once. If nothing needs trust, it does not ask.

Codex trust
  stop  needs trust  codex  /home/me/.codex/hooks.json:stop:1:0
  /usr/local/bin/notify-done

? Trust 1 hook in Codex?  Yes  No

After you confirm:

Codex trust
✓ stop  trusted  codex  /home/me/.codex/hooks.json:stop:1:0

--yes (-y) skips the question. Codex's review exists so that a person sees the command before it runs. Without a terminal and without --yes, nothing is trusted: the command lists the hooks and exits with an error that names --yes. --json never asks either: without --yes it reports the hooks as needed and exits non-zero, with no error text. --yes is only valid with --trust. The --json output has a codexTrust list. Each item has target, entry, key, command, action (needed, trusted or unchanged), and disabled when true.

Only trusted_hash is written, and only ever added. A hook you disabled in Codex stays disabled and is reported. enabled, the rest of config.toml and hooks Skillshare does not own are left alone. When a hook moves or is removed, its old entry stays in config.toml. It matches only a hook with the same event, matcher and handler at that position, so it cannot trust a different hook.

If codex is not on PATH or Codex fails, the hook files are still written and the command fails with "hook files were written, but Codex trust was not recorded". With several Codex homes, a home that Codex cannot list, or whose write fails, does not stop the others: the error names the homes that were recorded and the ones that were not. Plain hooks sync never starts Codex and never changes trust. --trust works in global mode only; sync hooks, hooks list and project mode reject it. The dashboard's Hooks page (global scope) has the same action: it lists the hooks that need trust with their commands and records trust only after you confirm in the dialog.

Git hooks

Git bindings use commands, with friendly names distinct from Git event names:

bindings:
  git:
    commands:
      project.check:
        events: [pre-commit]
        command: "{files}/check.sh"
        parallel: false
    files:
      check.sh: |
        #!/bin/sh
        exec make check

The dashboard edits the complete Git binding as YAML. Each command needs a nonempty list of unique, lowercase event names and a single-line command. Unknown events warn. Friendly names use letters, digits, _, - and ., begin with a letter or digit, have at most 128 characters, and cannot contain .., end in ., or equal a Git event name. Names must be unique across enabled entries sharing a destination. Use distinct friendly names for global and project commands too: Git merges both scopes. parallel needs Git 2.55+; older Git produces a warning.

Git appends the event's arguments to each command and gives each hook the full stdin. Put compound shell logic in files: commands ending in fi, done, esac, }, or ) are rejected because argument appending breaks them. {files} expands to a quoted absolute helper directory, requires files, and is regenerated for each machine. Helpers are UTF-8, normalized to LF and written executable. Commands retain native Git semantics and do not have Agent timeouts. Ordering is deterministic: entry names first, then friendly names; Git controls execution and parallelism. Skillshare leaves hook.jobs settings alone.

Includes and activation

Global outputs use $XDG_CONFIG_HOME/git/skillshare/ (default ~/.config/git/skillshare/). The include is added to GIT_CONFIG_GLOBAL when set, otherwise an existing ~/.gitconfig, otherwise an existing $XDG_CONFIG_HOME/git/config, otherwise a new ~/.gitconfig. GIT_CONFIG_GLOBAL must be absolute. Includes inside HOME use ~/ paths.

Project outputs and helpers live under Git's common directory. The include is skillshare/hooks.gitconfig in its config file. Linked worktrees share this destination; declare one root per common directory. Roots must be non-bare repository top levels. Unavailable roots or a missing Git executable skip Git outputs and retain their ownership records for a later retry.

Skillshare edits regular, writable include targets using Git's native lock. It never writes through a symlink. An absent include in a symlinked or unwritable target yields inactive with the exact lines to add manually:

[include]
    path = ~/.config/git/skillshare/hooks.gitconfig

Generated files can still be written while inactive. Git below 2.54 also reports inactive: there is no fallback dispatcher. Plans keep the actual add, update, remove or unchanged action and report the activation warning separately in inactiveReason. After sync, inactive outputs do not remain pending writes. Manual includes stay user-owned, including conditional includeIf lines; Skillshare never adds an unconditional include to widen their scope, including nested includes beneath inactive conditions. Unreadable or excessively nested include declarations fail closed. Global conditional activation is unknown until Git evaluates it in a repository. Project activation is checked from effective config origins. A present include does not alone prove activation. hooks list -g --json exposes the git capability/include object and projectGit for declared roots, including manual lines and core.hooksPath.

Conflicts, removal and migration

The generated hooks.gitconfig is owned as a whole file. Unmanaged or externally edited content conflicts even if identical. --replace explicitly adopts identical content or regenerates the whole file, discarding outside edits after backup. Active ownership by another config still blocks replacement. Friendly name collisions check command/event definitions across effective scopes and includes. Replacement can remove only that name's section in the writable regular target config; system or included-file collisions remain conflicts. enabled=false overrides warn rather than being overwritten. Previews and section backups expose only the affected hooks/include values, preserving unrelated private config.

Disable or remove with --sync removes owned commands, helpers and includes, then prunes only directories Skillshare created. Manual includes remain. Backups can restore the whole generated file, helpers or a narrow config operation while preserving unrelated later edits. Source definitions do not change during restore. remove --keep-files is refused for any Git binding: commands share one generated file, so detaching an entry is ambiguous.

Git config hooks coexist with hooks-directory scripts. Inspect existing core.hooksPath and hooks-directory registrations before migrating a tool; config hooks run before the hooks-directory hook and duplicates can run twice. Removing old copies is a separate user action. Git import, hooks-directory recognizers, script-mode bindings, the structured Git editor and doctor checks are reserved for later phases. This implementation supports config commands with inline helpers only. POSIX shell quoting and Windows-style forward-slash paths are rendered; native Windows hook execution is unverified.

See the Roborev recipe for portable wrappers.

Projects, conflicts and recovery

In a global config, hooks.projects maps absolute project roots to an entries mapping using the same Entry format. Manage these in Projects → Hooks. A project with its own .skillshare/config.yaml must be managed in project scope. Project-only sync applies that project's hook changes.

Sync preserves unrelated settings and unowned hooks. Matching content alone does not establish ownership. Modified owned outputs are conflicts, including on disable, remove and restore. Use the preview to inspect exact actions; an explicit replacement applies only to the selected entry. For a shared file, each plan line lists the events the entry adds (+), updates (~) and removes (−), also available as events in JSON plans. update means the entry keeps registrations in that file; remove means it leaves the file entirely. Edits keep the file's style, compact or indented, and a hooks key Skillshare added is removed again when its last hook leaves. A file Skillshare created is deleted once nothing else is left in it, together with the folders it created that are now empty; the preview shows this as a file deletion. Files and folders that existed before, and anything you added, stay. An externally edited output that can no longer be identified safely is released from ownership and left untouched; inspect the preview before publishing another registration.

Backups restore native outputs while preserving unrelated later edits; they do not replace your source definition. Use Settings → Backups → Hooks or hooks restore to preview and restore one backup.

Synced describes configuration written by Skillshare. Restart or reload the Agent as its native workflow requires; native trust, enabling hooks and code compatibility remain controlled by that Agent. Skillshare never executes hook commands during management or changes native trust automatically; only --trust records Codex trust, when you ask for it.

OpenCode v2 directory packages

On Windows, the hook version probe resolves npm's standard native .cmd shims for opencode-ai (later v1) and @opencode/cli (v2) directly to their installed bin/opencode.exe, without a shell. Older v1 Node-based launchers and custom batch wrappers are not inferred or executed. For those installations, set the OpenCode account's cli to the absolute path of its installed native opencode.exe (for example, the matching opencode-windows-x64-baseline package's bin/opencode.exe), or put that native executable's directory first on PATH. Skillshare does not run postinstall scripts or repair missing binaries.

Skillshare probes only opencode --version (or the account's cli). When the version cannot be detected, because the binary is missing or fails or reports a major version other than 1 or 2, only the OpenCode bindings are skipped: the preview shows a warning, existing OpenCode output is left as it is, and every other Agent syncs normally. v1 keeps the flat .ts file. v2 writes plugins/skillshare-NAME/index.ts plus package.json, declaring @opencode/plugin: ^2.0.0, and adds the directory's file:// URL to plugins in opencode.json(c). Project hooks use .opencode/opencode.json(c). Comments and unrelated registrations remain unchanged. Conflicting JSON/JSONC files or OPENCODE_CONFIG/OPENCODE_CONFIG_CONTENT overrides are refused for hooks. A code binding is written as given on either version; Skillshare does not check that it matches the v1 or v2 plugin API.

For a complete v2 package, use bindings.opencode.package instead of code/codeFile: a map of relative filenames to UTF-8 contents, including package.json. The manifest and its dependency requirements are preserved verbatim. Nested files are allowed; traversal, symlinks, file/directory collisions, .git, and node_modules are not package payloads. The native editor displays this file map as JSON. The server entry must be a string root or ./server export with a package name, or a conventional root server/index file (.ts, .tsx, .js). Conditional and fallback server exports are refused rather than guessed. A package binding needs OpenCode v2; on v1 it is skipped with a warning.

hooks import --from opencode also reads immediate package directories under the native plugins folder. Like flat code imports, it copies them to Skillshare's named output and warns you to remove the original files and registration after syncing to avoid duplicate loading. Installed node_modules and .git are not imported. A package holding a file that cannot be read, such as a symlink, is listed with that problem; the other packages and flat files stay importable. On OpenCode v1, directories under plugins are not listed as hooks. Sync, edits and removal use the same ownership checks and backups as other hooks; an externally edited package file conflicts rather than being overwritten.

Provision declared dependencies yourself before loading; the preview says so whenever it writes package files. Skillshare never runs plugin code, installs dependencies, or converts v1 code to v2. Registration is not proof the plugin loaded. The dashboard intentionally supplies no OpenCode starter code because the two APIs differ.