bstack full-discipline operating mode. Replaces the daily ritual sentence ("do this, document, be autonomous, ensure CICD checks green, work with parallel agents, think deeply through chain of dependencies, address PR comments, update docs, complete autonomously") with one invocation.
npx skills add broomva/autonomousThen in your agent session, invoke when the user has agreed on a plan and wants execution:
- Explicit:
/autonomous - Bare directives: "go", "proceed", "be autonomous", "merge autonomously", "automerge", "all green"
- Selection from options: "yes, option A", "let's do that"
When invoked, the agent runs a 20-step pipeline that fires every bstack reflex without further prompting:
- Pre-flight: state snapshot (P15), dep-chain trace (P14), worktree decision (P10), validation plan (P11), long-horizon check (P12)
- Plan: Linear ticket (P3), parallel decomposition (P5)
- Execution: empirical watchers (P11), research-before-implement (Context7), docs before push
- Pre-push: smoke tests, bookkeeping reflex (P6)
- PR + merge: PR with Linear ID (P4), CI watcher (P7), PR comment loop, auto-merge when gates pass
- Post-merge: janitor (P9, P10), dogfood receipt (P11)
This skill is the workspace-specific operationalization of the canonical autonomous-senior-engineer prompt. The prompt is universal (any repo); this skill grounds it in the bstack primitives of a bstack-enabled workspace.
The prompt sets the agent's stance. The skill says how that stance gets executed with these primitives. Together they're the discipline.
The user invokes
/autonomousto stop instructing the agent on bstack discipline. Asking them "should I check git status? open the PR? auto-merge?" violates the contract. Disciplines are unconditional defaults.
SKILL.md— full skill content (role contract, 20-step pipeline, role-contract ↔ bstack mapping, anti-rationalization table A + dump-extracted table B, red flags, output contract)https://broomva.tech/prompts/autonomous-senior-engineer— canonical universal role contract (embedded into SKILL.md, version-tracked)~/broomva/AGENTS.md§Bstack Core Automation Primitives — the P1-P16 reference this skill composes with
This skill follows the TDD-for-skills doctrine from superpowers:writing-skills. Every SKILL.md change must verify against the pressure scenarios in tests/pressure-scenarios.md:
- Scenario 1 — Moderate: single-vector pressure (
ship /api/health) - Scenario 2 — Stacked: time + sunk cost + authority + local-verified (
3h debug, hotfix-merge, 20min until standup) - Scenario 3 — Cross-repo trigger: inverse-section concrete machine-checkable test
- Scenario 4 — Public-API-break: inverse-section AST-diff test
Loopholes don't get accepted; they get closed. See CONTRIBUTING.md for the verification protocol.
We welcome contributions. The bar is high — every change is a discipline change that future agents will be bound by — but the process is approachable. See CONTRIBUTING.md.
Quick rules:
- All changes to
SKILL.mdmust pass the pressure-scenario verification - New anti-rationalization rows need the scenario that surfaced them (rule-of-three: ≥3 instances before crystallizing)
- The direction of evolution is stricter, not looser
This is a markdown skill — no executable code, no network calls. The trust boundary is the agent host, not the skill itself. Report security concerns per SECURITY.md.
See CHANGELOG.md. Notable releases:
- v0.0.2 (2026-05-13) — 20-reflex pipeline; 4 rationalization surfaces closed; stacked-pressure test scenario added
- v0.0.1 (2026-05-12) — Initial release; 19-reflex pipeline; embedded role contract from broomva.tech/prompts
broomva/bstack— the substrate this skill operates on (16 primitives + 29 skills)broomva/persist— bstack P12, long-horizon loop disciplinebroomva/p9— bstack P7, productive-wait optimizerbroomva/bookkeeping— bstack P6, knowledge graph maintenancebroomva.tech/prompts/autonomous-senior-engineer— the universal role contract this skill operationalizes
MIT — see LICENSE.