goal-acceptanceSAFE

Acceptance-criteria-driven goal completion for autonomous agents — core library, MCP server, and Cordis plugin

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
# @cckyros/goal-acceptance:skill
name: goal-acceptance
description: Acceptance-criteria-driven goal completion for the goal-acceptance plugin — lock criteria before implementing, track task status, validate with real evidence, and confirm by an independent reviewer. Trigger words: @cckyros/goal-acceptance, acceptance criteria, goal acceptance, goal planning
allowed-tools: Bash Read
---

# goal-acceptance — goal acceptance skill

This plugin ships 18 MCP/agent tools that drive acceptance-criteria-based
goal completion: you lock criteria before implementing, track task progress,
validate each criterion with real evidence, and block completion until an
independent reviewer confirms the work.

## When to use

- the user gives a multi-step task (implement / fix / refactor) with a
  verifiable outcome — start with the `goal-planning` companion skill
- the task requires criteria to be set, updated, validated, or confirmed
- a goal needs a task plan, task status tracking, or a completion check

## Workflow

1. **Plan** — spawn a planning subagent (see `goal-planning` skill) and
   lock criteria via `set_acceptance_criteria` BEFORE writing code.
2. **Decompose** — set the task plan with `set_task_plan`; update status
   with `update_task_status` as work proceeds.
3. **Validate** — after each task, run the real verification and record it
   with `validate_criterion` (evidence from actual execution, never memory).
4. **Confirm** — have an independent reviewer agent re-verify self-claimed
   passes with `confirm_criterion` (fresh evidence, high-confidence type).
5. **Complete** — only when `can_complete_goal` returns `allowed: true`.

## Orchestrated task loop (worker + reviewer subagents)

For task-granularity orchestration — an orchestrator dispatching work to a
worker agent and routing each finished task to an independent reviewer
subagent — use the two task nodes instead of free-form status updates:

1. **Dispatch** — `claim_task` (or `claim_task` with a task_id) marks the
   next dependency-ready task `in_progress` and returns it with its linked
   criteria: the task-granularity acceptance contract. The worker executes
   against exactly those criteria — workers do not validate.
2. **Review** — a reviewer subagent re-verifies the task and calls
   `review_task` with per-criterion verdicts (fresh evidence). Verdicts are
   formal — a single validation layer, no confirm step.
3. **Loop** — rejection keeps the task `in_progress`; `blockingCriteria`
   is the rework feedback. The worker fixes and the reviewer calls
   `review_task` again. On full pass the task completes and its dependents
   unlock — the response returns `nextTask` ready to claim.
4. **Gate** — `can_complete_goal` (or `goal-acceptance check` headlessly)
   still guards goal completion across all tasks.

Criteria link to tasks via `task_ids` at `set_acceptance_criteria` /
`amend_acceptance_criteria` time; link every criterion to the task that
produces its evidence.

## Companion skills

Eight per-tool companion skills ship with the plugin and activate on demand:

- `goal-planning` — decompose a task into coverage-complete criteria first
- `set-acceptance-criteria` — lock criteria (criterion quality rules)
- `amend-acceptance-criteria` — append criteria when requirements expand
- `get-acceptance-criteria` — read current criteria and progress
- `update-task-status` — track task status linked to criteria
- `validate-criterion` — record status + evidence from real execution
- `confirm-criterion` — independent reviewer confirmation (fresh evidence)
- `can-complete-goal` — completion gate; self-claimed passes block

## Config

- `pluginData` (env `PLUGIN_DATA`) — data directory; goals live in
  `<pluginData>/goals/`. Empty means in-memory mode.
- `autoSteerUncompleted` (dsh only, default true) — steer the agent back to
  uncompleted goals at turn end.
- `maxSteeringTurns` (dsh only, default 5) — steering budget per session.

Set values with `npx @cckyros/goal-acceptance config set <key> <value>` (project) or with
the `--global` flag. Env vars override file values.

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.