strategic-sweep

A downstream methodology package for constitutional-agent-governance.

The constitutional-agent-governance library evaluates agent decisions at runtime — the 6 Gates + 12 Hard Constraints that hold the floor. strategic-sweep is the layer above: on-demand strategic re-evaluation that pulls fresh external signal (regulatory, competitor, market, AI/agent governance, legacy reactivation), maps it onto 8 strategic frameworks (VRIO, Porter Five Forces, PESTEL, Blue Ocean, JTBD, Wardley, Crossing the Chasm, SWOT), and emits project-specific recommendations with falsification criteria + Cynefin classification + reversibility annotation.

The gates catch agent failures. The sweep catches strategic drift.

---

What this repo contains

• skill/strategic-sweep.md — the slash-command skill body. Designed for use with Claude Code (or any agent harness that supports prompted skills); adaptable to other orchestrators.
• templates/ — 9 reusable framework templates:
  • STRATEGIC_SWEEP.md — the sweep artifact template (frontmatter + 8 body sections + per-rec structure)
  • VRIO.md, PORTER_5F.md, PESTEL.md, BLUE_OCEAN.md, JTBD.md, WARDLEY.md, CHASM.md, SWOT.md — one per framework
• rules/verification_discipline.md — R-V1 through R-V6 verification rules + R-V7 proposal. Codifies the HC-8 fact-discipline that prevents sweep-output overstatement.

---

Methodology overview

A sweep runs as 5 phases:

1. Read canon — parallel-read existing framework artifacts + prior sweep + production gate + outreach state + memory. No external calls.
2. Pull signal — 10 WebSearch queries across 5 buckets (regulatory / competitor / market / AI-agent governance / legacy reactivation). Capture source URL + date + 1-line summary + confidence per result.
3. Map signals → framework deltas — every signal cites which of the 8 frameworks it touches and in which direction.
4. Compute VRI delta — Vision Realization Index = 0.10·inevitability + 0.10·co_option + 0.10·leapfrog + 0.70·market_validation. Pre-revenue cap = 70.
5. Emit recommendations — max 9 (3 IMMEDIATE / 3 TACTICAL / 3 STRATEGIC). Each maps to a project task; each has falsification criteria; each has Cynefin regime + reversibility annotation.

Disconfirming evidence section is mandatory (≥3 items per sweep). If a sweep has no honest weaknesses, the sweep is wrong.

---

Adaptation

Templates contain project-specific path placeholders (<project>/, <your-project>) and example references. Replace with your own project's equivalents:

Placeholder	Replace with
<project>/	your repo root
<your-project>	your product name
docs/<project>/...	your strategic baseline docs
memory/ paths	your agent's persistent memory directory
docs/runbooks/TASK_QUEUE.md	your task tracker

The skill body assumes Claude Code conventions (.claude/commands/, Skill invocation, Bash + WebSearch + Read + Write tools). Adapt as needed for other orchestrators.

---

Relationship to constitutional-agent-governance

The two packages compose vertically:

┌─────────────────────────────────────────────────────────┐
│ strategic-sweep (this repo)                             │
│   On-demand: signal → framework deltas → VRI → RECs     │
│   Cadence: 7+ days between sweeps; signal-driven only   │
└──────────────────────────┬──────────────────────────────┘
                           │ emits framework events
                           ▼
┌─────────────────────────────────────────────────────────┐
│ constitutional-agent-governance                         │
│   Runtime: 6 gates + 12 HCs evaluate every decision     │
│   System states: RUN / THROTTLE / FREEZE / STOP         │
└─────────────────────────────────────────────────────────┘

A sweep's recommendations become inputs to the constitutional gate evaluation. Example: a TACTICAL recommendation that hasn't moved in 30 days raises the ConstitutionalGate's bug_recurrence_rate for the orchestrator. A sweep's vri_delta_vs_prior is observable signal that the EconomicGate's market_validation component can consume.

---

Status

Experimental v0.1 — a methodology derived from use within the HRAO-E reference environment; not independently validated outside that context. See the hrao_e_production_evidence example in the parent library for what the gates + sweep discipline have caught together.

Possible future work (no committed timeline):

• protocols/hc8_corrections.md — corrections artifact protocol when prior sweep claims are falsified
• protocols/retraction_ledger.md — append-only ledger pattern + closure discipline
• protocols/governance_windows.md — amendment / ratification window pattern
• examples/ — sanitized example sweep + addendum chain

---

License

MIT — see LICENSE.