Install any skill in seconds. Free to start, no credit card required.
Get Started Free →chuspeeism/manage-taskboard
Manage Codex Taskboard / e-taskboard work with taskctl.
Passed 9 of 15 test cases with the skill, and 5 of 15 without it.
and 7 more in the full report
7 fixed7 passed either way8 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
This benchmark measured an earlier version of the skill.
A newer attempt on gemini-3.6-flash on 10/2/2026 did not produce a publishable comparison (not comparable). The snapshot above is the previously measured run.
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
We are updating Taskboard issue TB-702 while running inside a Claude Code session with session ID 'claude-sess-987'. Which CLI flags should we supply to taskctl to record session traceability for Claude Code?
| Checked | Without skill | With skill |
|---|---|---|
| Passes --agent-platform claude to taskctl. | Missed | Passed |
| Passes --session-id claude-sess-987 to taskctl. | Passed | Passed |
The instructions your model reads when the skill loads.
This skill serves the local-first Codex Taskboard product, including its configured LAN and cloud services.
Apply the workflow below only to work explicitly targeting Codex Taskboard or already established as belonging to it in the conversation. An issue identifier, repository, or generic request to manage tasks, sync status, or add comments does not establish that scope. For GitHub, Phabricator, or another external tracker, use that system's tools and workflow; do not query, claim, or mirror its issues in Taskboard unless the user asks for that board operation. When the target is unclear, clarify it before running taskctl.
Within that scope, use taskctl for every project, issue, relation, and comment operation. Consume its JSON output. Use the exact issue identifier returned by the taskboard or supplied in the prompt. Never assume, derive, or rewrite an identifier prefix.
Open only the relevant section of references/cli.md when command syntax is needed.
taskctl binary and Taskboard URL supplied by the task or injected runtime. Do not replace them with a global CLI, the default port, or another board.& "$env:LOCALAPPDATA\Codex Taskboard\bin\taskctl.cmd" issue get ID --json in PowerShell. The packaged wrapper reads the active launcher runtime. If this packaged path is absent, stop and ask for the exact installed taskctl.cmd path; do not switch to a global CLI or guess the service URL.'/Applications/Codex Taskboard.app/Contents/Resources/bin/taskctl' issue get ID --json. Keep the single quotes because the path contains a space. The packaged wrapper reads the active launcher runtime; do not search the filesystem for another CLI or reconstruct the tokenized URL.taskctl issue get ID --json. The desktop app adds its packaged wrapper to the managed Codex PATH; do not search the filesystem for another CLI or reconstruct the tokenized URL.In this product, companion means the device-local loopback service used for cloud mode (Codex/Git/Skill/MCP, path mapping, Basic Auth proxy). Related names: local companion, loopback companion, CODEX_TASKBOARD_COMPANION_URL, cloud-companion.json, LOCAL_COMPANION_REQUIRED.
When writing Chinese, keep the English word or use 本地 companion / 本地配套服务 / 环回代理. Never translate as 伴侣 or invent 伴侣 API. Ordinary task/comment/attachment HTTP routes (/api/tasks, /api/comments, /api/attachments, …) are the Taskboard HTTP API (or local server API)—not “companion API”.
issue get and comment list. Also run attachment list --task. On the first handoff, omit --after and read the full results. Keep the separate nextCursor from each list. When the same task resumes, run issue get again, then pass each saved cursor to its matching list with --after so only new or modified entries are returned. Comment lists include the attachments on returned comments; use attachment list --comment with its own cursor when a known comment attachment list can grow. Read the description and latest comments before deciding whether to start. Treat comments as current requirements, including returned work. If they say to wait, not execute, or not start now, stop and report without changing the status.backlog as not approved for execution. Unless the user explicitly authorizes that issue, do not claim it, move it to another status, or perform task work; its assignee alone is not authorization. If work may start, claim it before reading code, downloading attachments, analyzing the implementation, or doing any other task work. Move a claimable todo to in_progress with its current version; do not continue until the move succeeds. If it is already in_progress, continue only when it is bound to the current conversation. Never move an issue claimed by another conversation.version is stale, run issue get and comment list again. Retry once with the latest version only when the issue is still a claimable todo, is not bound to another conversation, is not archived, and its description and latest comments are unchanged. If it was claimed, its status or requirements changed, it is archived, the service is unavailable, a permanent API error occurs, or the retry fails, stop and report. Never loop or take over another agent's claim.context current. Treat its project as a workspace match only when project.workspacePath is the current directory or one of its ancestors. An unmatched local project is the documented fallback, not proof that the requirement belongs in the global project. If the user named a target project or the working directory identifies one, run project list, select that exact project by id or name, and stop to ask if the result is ambiguous. Search existing project issues before creating one in that confirmed project, then pass its explicit id to issue create. Update a matching issue instead of creating a duplicate. Use the fallback only when the user explicitly wants the global project. Do not track trivial requests.in_review with its current version.done only after the user explicitly accepts it or asks to complete it. Use blocked when work cannot continue and canceled when it will not continue.taskctl project readme get [PROJECT_ID] to inspect project architecture, constraints, and conventions before planning or executing complex tasks.docs/ folder.test cases passed, vs 5 / 15 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.