Skip to content

v1.2.0 — RFL Reflection Protocol Step + Naming Schema

Choose a tag to compare

@ogerly ogerly released this 09 Apr 10:48
· 48 commits to main since this release

What's New

RFL — Reflection Protocol Step

The most common failure pattern in agentic sessions: Agent decides the opposite of Session 12 in Session 47 — without noticing. All four documentation layers store. None verify.

RFL closes this gap as a protocol step (not a layer) in on_session_start:

  1. Stage 1: Pattern-match on TOPIC tag in closed workpaper filenames (*-{TOPIC}-*)
  2. Stage 2: Use LTM query results to find related prior decisions
  3. Stage 3: Read the most recent closed workpaper (chronological fallback)

If a prior decision conflicts with the current session goal → flag in workpaper under ## ⚠ RFL Consistency Flag. No conflict → zero overhead.

AAMS is the first open standard to define explicit cross-session consistency checking.

Naming Schema

Structured filenames for workpapers and whitepapers:

  • Workpapers: {DATE}-{TOPIC}-{SUBTOPIC}-{description}.md
  • Whitepapers: WP-{NNN}-{TOPIC}-{description}.md
  • Topic Registry: ARCH, SPEC, LTM, SEC, BOOT, FLD, RES, MKT, ISS, GOV, EDU

TOPIC tags are the primary key for RFL pattern-matching. Schema is recommended (not enforced) — hybrid approach for real-world compliance.

SCIENCE → Framework

The Knowledge Validation Layer (SCIENCE) was designed as a fifth documentation layer and intentionally deferred to the Agent-Loop-Framework level. Spec = body (structure, storage, protocol). Framework = mind (reasoning, research, validation). Four layers remain.

Open Question

  • #40: Diary granularity — is monthly good enough?

Files Changed

  • READ-AGENT.md — Naming Schema section + RFL in session-start contract
  • .agent.json — Updated naming, RFL in on_session_start
  • WP-001 — Naming Schema + RFL as core elements
  • New workpapers, diary entry, LTM ingest (#093-#096)

Closes: #36, #37, #38, #39