All skills
boshu2 avatar

/craft-goal

@833aefb
by Boboshu2/agentops446 stars
42

Draft or lint a bounded persistent goal above a bead graph of RPI experiments. Use when: this goal workflow is explicitly selected; shaping a single change belongs to Plan.

Use this Skill: https://skilld.dev/gh/boshu2/agentops/craft-goal

This session only. Nothing lands on disk.

referencesgoal-prompt.md

≈831 tokens on demand. Your agent reads this file only when SKILL.md points to it.

Mayor-style goal prompt

Copy this prompt verbatim, replacing every angle-bracket field. Do not delete the wave, hard-envelope, and terminal-report sections.

Goal outcome:
<larger caller-visible result>

Terminal acceptance and evidence:
1. Given <state>, when <event>, then <observable result> — <authoritative proof>

Domain terms:
- <term>: <one-line definition>; use only this word in beads, code and tests

Non-goals and authority:
- <excluded outcomes/mechanisms>
- Reads/writes/external/Git authority: <exact scope>

Bead graph:
- Root epic/mol: <existing id or bounded bootstrap rule>
- Initial experiments, if known: <bead → criterion/uncertainty>
- Record notes, scratch, evidence, verdict refs, and dependency/provenance links.

Experiment policy:
- Apply the `navigate` skill each wave: observe, pick, ratchet, checkpoint.
  These policy lines bind with or without it.
- One bead is one RPI experiment.
- Select only work tied to an unmet criterion or named blocking uncertainty.
- Consume each verdict unchanged; useful progress needs evidence tied to an
  unmet criterion or a blocking uncertainty, not digest/count movement alone.
- Distinguish pre-existing discovery from introduced regression using causal
  evidence; unknown cause and recurrence require HOLD, not a design diagnosis.
- Classify discoveries as necessary-now, linked-follow-up, or HOLD/rescope.
  Never downgrade a necessary finding to optional to obtain completion.
- Retain evidence/provenance; revise or withdraw beliefs when evidence changes.

Wave envelope:
- <numeric RPI/concurrency/time/token/live-attempt limits>

Hard goal envelope:
- <numeric total RPI/time/token/live-attempt/compaction/surface limits>
- No artifact, repair, helper, subject, compaction, or wave resets a total.
- Include helper and validation costs inside the allowance.
- Enforcing native controls and observable remaining allowance: <actual controls
  and measurements; explicitly identify any unmeasured aggregate>.
- Objective text alone does not enforce a budget or a native pause.

Breaker and andon:
- Ordinary informative red may produce a materially different next experiment.
- <threshold> non-ratcheting results, oscillation, regression, unknown defect
  cause, recurrence, or scope pressure: HOLD implementation for causal review.
- Consult exactly one bounded fresh helper per HOLD incident inside the existing
  allowance; repeated continuation of that incident does not reset the helper.
- UNSTUCK names a different admissible experiment and discriminating check;
  ESCALATE or no useful admissible experiment reports NEEDS_OPERATOR.
- Cancellation, explicit refusal/judgment, or spent hard time/cost/quota skips
  the helper and stops work; a retry threshold alone is not a spent budget.

Wave checkpoint:
- acceptance matrix; graph frontier; verdict/evidence summary;
- ratchets versus non-progress; measured remaining budgets; next thesis;
- observed native continuation/stop state, outstanding gaps, and helper use.
  Never claim a native pause or aggregate enforcement based only on this text.

Terminal reports:
- ACHIEVED: every terminal criterion is proven.
- NOT_ACHIEVED: envelope or permitted search is exhausted; report exact gaps.
- NEEDS_OPERATOR: judgment, rescope, or helper escalation; stop implementation.

Source: SKILL.md on GitHub

No alerts3d3 checks · Risk SAFE
  • Gen Agent Trust Hub3d

    The skill drafts and reviews autonomy contracts for agent experiments. It is susceptible to indirect prompt injection because it incorporates untrusted user data into a generated prompt that governs downstream agent behavior. It also executes local validation scripts using computed paths.

  • Socket3d

    No alerts

  • Snyk3d

    Risk: LOW · No issues

Signed by skilld at 833aefb. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 20 hours ago.

Activeupdated last week

README badge

README badge for boshu2/agentops/craft-goal