codex-goal-profile
SkillProductivityUse when converting an Arcanum work-pack task or SWU into a strong native Codex /goal command with outcome, verification surface, constraints, boundaries, iteration policy, and blocked stop condition.
Available today. Use it from your connected AI after setup.
No other account needed.
Connect ahel once, and every AI you use reads what you have installed.
Then ask your AI: use the codex-goal-profile skill
What this skill tells your AI
The instructions your AI receives, as published by cyberalchemyai/arcanum in .claude/skills/codex-goal-profile/SKILL.md and read by ahel’s review.
- a work-pack, task, SWU, or explicitly selected one-shot stream is ready for execution,
- native Codex Goals are available,
- the task may require continuation across turns,
- the user needs a compact, auditable
/goalcommand rather than another planning artifact. - a richer Arcanum execution frame must be preserved through sidecar context instead of exceeding the native goal text budget.
- work-pack path,
- selected task ID, SWU ID, or explicit one-shot stream ID,
- parent task contract,
- dependencies,
- source contracts,
- write scope,
- done criteria,
- validation command or evidence surface,
- handoff pack Markdown path,
- handoff pack JSON/index path,
- strict coverage status,
- fallback exploration rule,
- blocker state,
- budget or stop constraints.
- native goal character budget, default
4000, - optional private/runtime decision profile path, for example
.arcanum/profiles/decision-profile.yml, - optional capability policy for one-shot runs: allowed sigils, subagent authorization, receipt expectations, and stop gates.
When the full execution vision does not fit:
- Write or reference a sidecar profile/handoff artifact that carries the richer Arcanum context, capability policy, work-pack navigation, and verification details.
- Keep the
/goalline compact: name the outcome, verification surface, source sidecar, write boundary, iteration rule, and blocked stop condition. - Treat over-budget native goal text as
blockunless a sidecar-backed compact rewrite is produced.
Do not stuff the goal with every source excerpt. The native goal should point to the selected evidence pack and sidecar; the pack carries density, the goal carries control.
- risk posture and default-deny behavior,
- approval and mutation gates,
- slice size and sequencing,
- ownership boundaries,
- gap-filling and interrogation style,
- technique preferences,
- anti-pattern stop conditions.
The generated profile may summarize only operational effects needed for the current goal. Do not copy private profile contents into reusable public examples, canonical skill text, or public package fixtures. Record the profile path and the policy fields consumed in the audit block.
- run prerequisite/contract units before runtime units,
- stop at blocker-level decisions and route them through
decision-gate, - use
refineonly to repair or re-scope named design/plan gaps, - use
invokeonly to author missing define/design/plan/handoff artifacts named by the pack, - use
craftonly to update a scoped project ledger when the pack/profile asks for durable blocker, decision, gap, or next-move state, - spawn subagents only when explicitly authorized by the work-pack, dispatch, sidecar profile, or operator instruction, and require role receipts before parent synthesis,
- keep Task Session or Codex runtime evidence separate from reusable sigil validation evidence.
The one-shot goal should empower Codex to finish the selected stream, not to silently widen project scope.
- can be pasted directly as a native Codex
/goalwithin the configured character budget, - names the exact task, SWU, or explicit one-shot stream,
- has a measurable completion condition,
- names the verification surface,
- constrains write scope,
- references the session-evidence handoff pack and JSON/index,
- requires pack-first execution,
- permits broad exploration only for named gaps from the pack,
- requires reporting extra sources used for named-gap fallback exploration,
- summarizes any decision-profile policy consumed without leaking private profile contents,
- gates one-shot subagent and sigil use through explicit capability policy and receipts,
- names dependencies and blockers,
- explains what Codex should do between iterations,
- states when to stop and what to report.
- creating an Arcanum
/goalcommand that competes with native Codex Goals, - generating a goal for an unselected task bundle,
- generating an over-budget goal instead of compacting or using a sidecar,
- copying private decision-profile contents into public reusable artifacts,
- treating a decision profile as permission to ignore explicit human gates,
- authorizing subagents,
refine,invoke,craft, ordecision-gateas broad ambient authority instead of bounded capability lanes, - omitting verification,
- hiding blockers,
- allowing broad write scope by default,
- generating a goal without strict handoff-pack coverage,
- treating fallback exploration as permission for broad discovery,
- saying "keep going until done" without a budget or stop condition,
- marking the goal complete without evidence.
## Codex Goal Profile Result
- Source work-pack: <path>
- Selected unit: <task-or-swu-id>
- Readiness: pass | block
- Goal budget: <limit and pass | block>
- Decision profile: <path or none; consumed fields or n/a>
- One-shot mode: yes | no
- Capability policy: <allowed sigils/subagents and receipt gates or none>
- Sidecar profile: <path or n/a>
- Native Goal:
```text
/goal <goal text>
- Verification surface:
- Boundaries:
- Handoff pack: <markdown path and JSON/index path>
- Strict coverage: pass | block
- Fallback exploration: none | named gaps only | block
- Extra-source reporting: required | n/a
- Stop condition:
- Validation:
</output-contract>
Signals
- GitHub stars
- 25
- Forks
- 3
- Last commit
- Sep 2026
Advanced
- Catalog kind
- skill
- Gateway key
codex-goal-profile- Source
- github.com/cyberalchemyai/arcanum