Skip to main content
Status: active · Version 1 · Filed 2026-04-17

​
Plan 2: Methodology Enforcement + Cross-Agent Continuity

Source: Donna’s second review session on 2026-04-17 (Lola project, claude.ai). Builds on Plan #1.

​
Goal

Operationalize the four-ring layered enforcement architecture (ADR #6). Make methodology compliance emerge from utility. Enable cross-agent continuity through the hd_exec pickup verb. Extend Prism beyond code projects via ptype expansion.

​
Parent ADRs

  • ADR #6 — Layered enforcement architecture (4 rings)
  • ADR #7 — PRISM.md as composed template with opt-in upgrades
  • ADR #8 — hd_exec verb for symmetric handoff pickup
  • ADR #9 — Ring 4 soft enforcement with telemetry
  • ADR #10 — Ptype expansion to research / travel / home-improvement

​
Sequencing

​
Wave A — The BIOS and the constitution (do first)

  1. Rewrite CLAUDE.md as thin BIOS. Target: fewer than 20 lines. Imperative launch sequence: (a) read AGENTS.md, (b) read PRISM.md if present, (c) call prism_start with PID from PRISM.md header or project_dir lookup, (d) display status card. No methodology content in CLAUDE.md itself — only the launch imperative.
  2. Rewrite AGENTS.md as the universal constitution. Move all truly-universal principles here (Context Is Currency, Explicit Over Implicit, AI First/Only, never delete, session lifecycle). Everything code-specific moves to prism-application.md in Wave B.
  3. Create the Prism install directory structure for templates: templates/prism-base.md, templates/prism-application.md. Stubs OK for now; populate during Wave B.

​
Wave B — The template system

  1. Author templates/prism-base.md. Universal methodology: session lifecycle (start/wrap/handoff), artifact vocabulary (Plan/Spec/Decision/TODO/Journal/Retro/Note/Memory), what each is for, when to use which.
  2. Author templates/prism-application.md. Code-project specifics: ADRs for architectural decisions, plans for phases, retros at phase boundaries, specs as requirements, tests-before-code discipline.
  3. Extend prism_scaffold / bootstrap_project to compose PRISM.md from prism-base.md + prism-{ptype}.md at bootstrap time. Write methodology_version header into generated PRISM.md.
  4. Backfill: generate PRISM.md for PID-PGR01 from current templates as a smoke test.

​
Wave C — hd_exec verb

  1. Implement hd_exec in Prism backend. Resolve PID from project_dir or prompt. Fetch last handoff doc + recent deltas + focus-tagged TODOs + recent ADRs + current-phase retros. Return structured payload with status_card.
  2. Add session marker: hd_exec sets a session_source=‘handoff_pickup’ flag that prism_wrap can inspect later (enables Ring 4 enforcement specific to handoff pickups).
  3. Expose hd_exec as an MCP tool in the Prism MCP server. Add to tool_search inventory with description clarifying it is symmetric to prism_wrap (wrap closes, hd_exec opens).
  4. Document hd_exec usage pattern in AGENTS.md and prism-base.md: ‘When starting a session to continue another agent’s work, call hd_exec instead of prism_start.‘

​
Wave D — Ring 4 enforcement hooks

  1. Implement prism_start rules_reminders array. Server computes: N open TODOs, last wrap timestamp, phases without retros, methodology_version vs canonical, TODOs tagged to last handoff focus.
  2. Implement prism_wrap accounting inspection. Server checks session for prism_todo complete calls, prism_journal writes, prism_retro creates, prism_decide filings. Returns warning payload (not block) if work session had no accounting.
  3. Add telemetry: per-agent accounting-completeness rate. Simple counter table: (agent_id, wraps_total, wraps_with_journal, wraps_with_todo_close).
  4. Add methodology-version drift detection in prism_start. When PRISM.md methodology_version is behind canonical template version, return upgrade prompt in rules_reminders.

​
Wave E — Ptype expansion

  1. Author templates/prism-research.md overlay. Literature review as a phase, specs as claims-with-citations, decisions as methodology choices, TODOs as experiments/sources, retros at draft milestones. Lean on graph leg (Neo4j) for citation relationships.
  2. Author templates/prism-travel.md overlay. Days as phases (or legs as sub-projects via parent_pid), bookings as ADRs, contacts as notes, itinerary as plan.
  3. Author templates/prism-home-improvement.md overlay. Research phase before action, vendor/product decisions as ADRs, failure-mode notes as G-scope candidates.
  4. Smoke test: bootstrap a travel project (—ptype=travel). Frank runs through a real trip planning session. Retro afterward captures what felt natural vs forced — direct input for methodology refinement.

​
Out of Scope

  • Hard enforcement (blocking prism_wrap on missing accounting). Explicitly rejected by ADR #9.
  • Auto-upgrade of PRISM.md without user consent. Explicitly rejected by ADR #7.
  • Migrating existing projects’ CLAUDE.md / AGENTS.md — those can stay as-is until Frank chooses to adopt the thin-BIOS pattern. PID-PGR01 is the pilot.

​
Success Criteria

  • New project bootstrap produces CLAUDE.md + AGENTS.md + PRISM.md with correct ptype overlay.
  • hd_exec on PID-PGR01 returns Donna’s handoff doc + all her open work in a single structured payload.
  • prism_start on a project with unaccounted sprints produces visible drift warnings.
  • Travel ptype smoke test completes end-to-end and the retro provides real methodology feedback.

​
Source Artifacts

  • ADRs #6 through #10 on PID-PGR01.
  • This plan supersedes none of Plan #1’s waves; they run in parallel.
Last modified on April 20, 2026