From codex-next
Executes or resumes a local goal plan read-only while tracking status in goal-tasks.md and logging evidence in goal-log.md. Useful for recoverable long-running tasks.
How this skill is triggered — by the user, by Claude, or both
Slash command
/codex-next:core-goal-runThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use this skill when the user provides a local goal or plan file and asks to
Use this skill when the user provides a local goal or plan file and asks to track completion, continue execution, skip human-only items, or leave a handoff.
Typical source file names include:
goal-plan-list.mdcurrent-plan.mdtodo.mdKeep long-running Codex work recoverable without turning it into a workflow platform.
The goal directory uses three plain Markdown files:
<goal-dir>/
├── goal-plan-list.md # original input, read-only reference
├── goal-tasks.md # current task list and status truth
└── goal-log.md # execution notes, evidence, and handoff
If the source file is not named goal-plan-list.md, keep its original name and
still create goal-tasks.md and goal-log.md beside it.
Do not create .goal-run/, run-id directories, scripts, hooks, registries, or
automation unless the user explicitly asks for a heavier tracker.
Treat the original goal file as read-only source context after
goal-tasks.md exists.
Do not keep updating the original goal file as the status tracker. If it
contains old status notes, treat them as historical input and reconcile the
current state into goal-tasks.md.
goal-tasks.mdThis is the current status truth. Keep it short.
It should answer:
Use this compact structure:
# Goal Tasks
Source: `goal-plan-list.md`
Status truth: this file.
Execution log and evidence: `goal-log.md`
Allowed status:
- TODO
- DOING
- VERIFYING
- DONE
- BLOCKED
- HUMAN_PENDING
- SKIPPED_HUMAN
| ID | Source | Task | Status | Auto | Evidence | Notes |
|---|---|---|---|---|---|---|
| T001 | goal-plan-list.md §47 | Review current state | DONE | yes | goal-log.md#t001 | current authority confirmed |
| T002 | goal-plan-list.md §8 | Implement frontend task X | DOING | yes | pending | small reversible diff |
| T003 | goal-plan-list.md §12 | Manual credential validation | SKIPPED_HUMAN | no | goal-log.md#t003 | requires credential |
Do not put long command output, long rationale, PR bodies, screenshots, or full
logs in goal-tasks.md.
goal-log.mdThis is the execution record, evidence index, and human handoff.
It should answer:
Use simple sections:
# Goal Log
Source: `goal-plan-list.md`
Task status truth: `goal-tasks.md`
## YYYY-MM-DD Run
Request:
- Continue automatic tasks from `goal-plan-list.md`.
- Skip human-only work.
## T001
Status: DONE
Evidence:
- Reviewed `goal-plan-list.md`.
- Confirmed current authority and stale context boundaries.
Notes:
- Older local directories are historical only.
## Handoff
Read first:
- `goal-tasks.md`
- `goal-log.md`
Next:
- Continue T002.
- Keep T003 as `SKIPPED_HUMAN` unless the user provides credentials.
Use only these statuses unless the user explicitly asks for a different local taxonomy:
TODO: known task, not started.DOING: currently being worked.VERIFYING: implementation or artifact is ready, proof is still running or incomplete.DONE: completed with evidence in goal-log.md.BLOCKED: cannot continue without a concrete blocker being resolved.HUMAN_PENDING: waits for human decision, credential, approval, or external action.SKIPPED_HUMAN: intentionally skipped because it is human-only and the user asked to skip human work.Do not invent compound statuses such as DONE_BUT_NEEDS_CI. Put that detail in
Notes or goal-log.md.
goal-tasks.md and goal-log.md if missing.HUMAN_PENDING or SKIPPED_HUMAN; do not mark them
DONE unless the user actually completed the human action and evidence is
available.goal-tasks.md and append concise
evidence or handoff notes to goal-log.md.DONE task has evidence, every BLOCKED task
names the blocker, and the Handoff section states what to read and do next.Evidence can be concise. Prefer pointers over pasted output.
Good evidence:
Do not treat goal-tasks.md itself as proof of completion. It is a tracker, not
the evidence source.
Subagents may investigate, implement, test, or review scoped work. The main
thread should remain the single writer for goal-tasks.md and goal-log.md to
avoid conflicting status edits.
If a subagent produces evidence, summarize it in goal-log.md and update the
matching row in goal-tasks.md.
Stop and report clearly when:
goal-tasks.md contradict each other.npx claudepluginhub blueskyxn/codex-is-all-you-need --plugin codex-nextExecutes tasks from PLAN.md sequentially with human oversight, handling task splitting, clarifying questions, tests, and learning persistence. Useful for deliberate progress on planned work.
Creates file-based project plans and tracks progress across complex tasks using task_plan.md, findings.md, and progress.md. Activates on planning requests, multi-step work, or research requiring over 5 tool calls.
Tracks task completion in markdown plan files using plan-file-management subcommands. Mandates test verification of observables and evidence before marking tasks done. Activates on *-plan.md files.