vonbai/goalx/skill

GoalX

GoalX is the default operating system for durable autonomous work in this project. Use it when the user wants a real run with durable state, not a one-shot answer.

Use GoalX By Default

Prefer GoalX when the user wants:

• autonomous investigation
• autonomous implementation
• a repo audit with follow-through
• durable observation and redirection
• a stopped or stranded run recovered in place
• a saved run continued into the next phase
• continuous improvement over time

Do not default to manual tmux orchestration. Use GoalX commands and durable surfaces.

Write Better Goals

Write the desired end state, not a checklist.

Good:

goalx run "users can complete the core workflow reliably in production"
goalx run "the codebase has a high-quality architecture audit and a prioritized fix plan"
goalx run "this project keeps getting better until the budget runs out" --intent evolve --budget 8h

Bad:

goalx run "1. inspect auth 2. patch middleware 3. update tests"

When the user is vague, keep it outcome-focused:

• "make deploy work" -> goalx run "the project deploys cleanly to the target host in one step"
• "audit this repo" -> goalx run "full audit of this project with an actionable improvement plan"

Default Operator Loop

Use this by default:

goalx run "goal"
goalx run --objective "goal"
goalx run --objective-file /abs/path/to/objective.txt
goalx status
goalx observe
goalx context
goalx afford
goalx tell "redirect"
goalx verify
goalx result
goalx save

Quick meanings:

• status: compact durable control summary
• observe: full diagnostic path for live transport and runtime facts
• context: canonical identity, paths, cognition, and assurance facts
• afford: current run-scoped command surface
• tell: durable redirect to master or a worker
• verify: record assurance evidence, not completion
• save: export a saved run for continuation

Use goalx schema <surface> before goalx durable write <surface> ... when authoring machine-consumed state.

Operator guidance:

• use --objective-file for long or multi-line objectives
• fresh runs can briefly show launching while bootstrap settles
• in that startup window, prefer status, observe, or goalx wait --run RUN master --timeout 30s
• do not default to recover unless the run is actually stopped or stranded

Resource safety:

• GoalX records resource facts itself; do not read Linux telemetry by default
• healthy runs stay compact on status
• use observe when you actually need runtime/resource diagnosis
• unsafe new launches can be refused explicitly instead of silently degrading effort or fan-out

Canonical Surfaces

GoalX now centers these durable surfaces:

• objective-contract
• obligation-model
• assurance-plan
• evidence-log
• cognition-state
• impact-state
• freshness-state

The old goal and acceptance runtime path is gone. Saved-run continuation now fails fast if canonical surfaces are missing.

Intent Routing

goalx run "goal"
goalx run "goal" --intent explore
goalx run "goal" --intent explore --readonly
goalx run "goal" --intent evolve --budget 8h
goalx run --from RUN --intent debate
goalx run --from RUN --intent implement
goalx run --from RUN --intent explore

Intent meanings:

• deliver: default shipped-result path
• explore: evidence-first investigation
• evolve: continuous improvement
• debate: challenge and refine prior findings
• implement: build from prior evidence

--readonly declares a no-edit GoalX boundary in target.readonly.

Recovery vs Saved Continuation

Same run:

goalx recover --run RUN
goalx budget --run RUN --extend 2h

New phase:

goalx save --run RUN
goalx run --from RUN --intent debate
goalx run --from RUN --intent implement
goalx run --from RUN --intent explore

Rules:

• recover relaunches the same run in place
• recover is for stopped or stranded runs, not a fresh run still showing launching
• save + run --from creates a new phase
• exhausted-budget recovery requires changing budget first

Selection Policy

Normal engine/model policy lives in ~/.goalx/config.yaml under selection.

selection:
  disabled_targets:
    - claude-code/sonnet
  master_candidates:
    - codex/gpt-5.4
    - claude-code/opus
  worker_candidates:
    - codex/gpt-5.4
    - claude-code/opus
  master_effort: high
  worker_effort: high

Guidance:

• keep selection user-scoped
• do not invent project-scoped engine policy
• use explicit --engine/--model only for one-off overrides

Optional Repo Cognition

GoalX always has builtin repo-native cognition. GitNexus is optional.

Current GitNexus behavior:

• binary install is preferred
• install with npm install -g gitnexus@1.5.0
• verify with gitnexus status
• pinned npx gitnexus@1.5.0 is only exposed when a real probe succeeds
• GoalX does not auto-install it
• GoalX records provider facts per worktree scope
• available does not mean indexed or fresh
• GoalX can best-effort refresh missing or stale GitNexus indexes during lifecycle transitions
• both master and worker scopes can receive runnable GitNexus cognition commands through goalx afford

If the current runtime already supports MCP, GitNexus MCP can be configured separately and then preferred for graph reads when freshness is trusted.

Worktree And Merge Boundaries

goalx add --run NAME --worktree "task"
goalx keep --run NAME session-1
goalx integrate --run NAME --method partial_adopt --from run-root,session-2
goalx keep --run NAME

Meaning:

• keep session-N: merge a reviewed session branch into the run root
• integrate: record run-root integration lineage after master already merged
• keep without session-N: merge run root into source root

Advanced Control

Use references/advanced-control.md when the user explicitly wants the full command matrix or operator-level control.