RFC — Artifact Write Protocol
This RFC is a first pass at specifying what the Nest tells agents to write, where each artifact belongs, and how those artifact types relate to each other.
The immediate goal is not to finalize the protocol. It is to expose the current state clearly enough that we can resolve contradictions instead of continuing to operate on overlapping norms.
Objective
Define a vault-wide write protocol for:
- durable wiki notes
- fleeting notes
- handoff notes
- seams
- PoShWiKi session pages
- session logs
- file-local evidence and generated artifacts
The core question is:
> when an agent learns something, decides something, delegates something, or pauses work, what exact artifact should it write, and where?
Why This RFC Exists
The current Nest has strong building blocks but inconsistent guidance.
The ambiguity is not about whether the vault works. It does. The ambiguity is about whether the write behavior is:
- fully specified
- internally consistent
- predictable across agents
- distinguishable by artifact type and lifecycle
At the moment, the answer is "not yet."
Current Canonical Surfaces
Based on the current directives and actual practice, the Nest uses these write surfaces:
1. 01_Wiki/
Used for:
- permanent notes
- literature notes
- specs
- MOCs and bridge notes
- dedicated handoff notes
- some fleeting planning / execution notes
2. PoShWiKi Session Page
Used for:
Session GoalActionsCurrent SeamNext Steps
This is the main session-state surface described by inter-agent-handoff-protocol and implemented by New-WikiSeam, Invoke-WikiLog, and Invoke-WikiNote.
3. Structured SQLite Tables via PoShWiKi
Used for:
SeamsDebates
This is a machine-readable sidecar for specific procedural artifacts.
4. 02_System/log.md
Used as the durable public ledger of:
- notable sessions
- multi-step work summaries
- maintenance passes
- synthesis outcomes
- architectural changes
5. File-Local Artifacts
Used for:
- experiment outputs
- review reports
- raw evidence
- generated portal files
- workbench outputs
These usually live under:
04_Experiments/02_System/00_Raw/03_Web/public/
Documented Rules vs. Emergent Practice
A. Session Tracking Location Is Underspecified
visitor-directives says:
- durable knowledge belongs in
01_Wiki/ - active work/logs should use PoShWiKi
- if a seam is longer than one line, create a dedicated handoff note
That is directionally good, but it leaves open:
- when a dedicated handoff note is preferable to only a seam
- whether a fleeting planning note is different from a handoff note
- when a session summary belongs in
log.mdversus the session page versus both
B. Note Type Semantics Are Not Fully Aligned
Current sources disagree:
- visitor-directives describes note frontmatter minimally and does not mention the full live type vocabulary
- yanp-for-agentic-workflows describes a lifecycle and intent model
- wiki-as-codebase describes
mocas a type, but the actual validator does not usemocas a live allowed type audit-yanp.ps1allowscommunity,community-report,experiment,fleeting,handoff,literature,permanent, andspec
So the conceptual model and the enforced model have drifted.
C. Status Semantics Are Not Fully Aligned
Current sources disagree on status values:
- visitor-directives mentions
draft | active - agent-note-conventions mentions
draft,active,archivedplus workflow signals - yanp-for-agentic-workflows describes
raw → draft → active → archived - actual notes also use values such as
supersededandpartially-resolved
So the status field is operationally richer than the main protocol notes currently admit.
D. Authorship / Metadata Rules Drift
agent-note-conventions requires or strongly implies fields like:
hostnamestatus_log
But those are not consistently present across the actual vault and are not enforced by the current validator.
So we have a difference between:
- aspirational metadata discipline
- live enforced minimum
- actual fleet practice
E. Seam vs. Handoff vs. Log Summary Is Blurry
Right now:
- a Seam is a resume boundary with
Goal,Current Seam, andNext Step - a handoff note is a bounded brief for another agent
- the system log is the durable summary of what happened
But the thresholds between them are not explicit enough.
This causes repeated judgment calls about:
- whether to write only a seam
- whether to also write a handoff note
- whether to summarize in
log.mdnow or later
Working Problem Statement
The Nest currently lacks a single explicit artifact taxonomy that answers:
- what each artifact type is for
- where it lives
- whether it is durable or procedural
- who is expected to read it next
- when it should be promoted, summarized, archived, or discarded
That is the protocol gap this RFC is trying to close.
Draft Artifact Taxonomy
This is a proposed first cut, not yet the final model.
1. Permanent Note
Purpose:
- durable conceptual knowledge
- load-bearing synthesis
Location:
01_Wiki/
Reader:
- future agents and humans as a stable knowledge surface
2. Literature Note
Purpose:
- source-grounded summary of an external corpus
Location:
01_Wiki/
Reader:
- future synthesis passes and fact-checking work
3. Fleeting Work Note
Purpose:
- bounded planning, research packet, temporary execution scaffold, or intermediate synthesis artifact
Location:
01_Wiki/
Reader:
- current or near-future agents/humans
Lifecycle:
- should usually be promoted, archived, or retired once its value is absorbed
4. Handoff Note
Purpose:
- bounded work brief for another agent when a seam alone is too lossy
Location:
01_Wiki/
Reader:
- a specific next agent or the HITL
Constraint:
- facts, constraints, recommendations, and next decision should remain clearly separated
5. Seam
Purpose:
- the minimum viable resume record at session boundary
Location:
- PoShWiKi session page and
Seamstable
Reader:
- the next agent or the HITL resuming immediately after the current session
Constraint:
- must be short, exact, and execution-oriented
6. Session Log Entry
Purpose:
- durable human-readable summary of meaningful completed work
Location:
02_System/log.md
Reader:
- anyone reconstructing vault history or operational state
Constraint:
- summarizes completed action; does not replace the seam
7. File-Local Evidence Artifact
Purpose:
- raw output, generated evidence, review report, experiment result, or compiled artifact
Location:
- local artifact path under
00_Raw/,02_System/,03_Web/, or04_Experiments/
Reader:
- follow-on verification or reproduction work
Constraint:
- should be referenced by seam, handoff, log, or note when it matters
Draft Decision Rules
These are the rules I think we want, but they need review.
Rule 1: Session State Does Not Belong In Permanent Notes
If the content is mainly:
- what I just did
- what failed
- what to do next
- what environment quirk I discovered
then it belongs first in:
- PoShWiKi
Actions Current SeamNext Steps
not in a permanent note.
Rule 2: Seam Is Mandatory, Handoff Note Is Conditional
Always write a seam at session end.
Write a dedicated handoff note only when:
- the next agent needs more than a short resume boundary
- verified facts and recommendations must be preserved distinctly
- the environment or scope could be misunderstood without a structured brief
Rule 3: Log Is Summary, Not Working Memory
02_System/log.md should record:
- durable summaries of meaningful work
- not every operational step
- not unresolved live state that only makes sense in the immediate session
Rule 4: Fleeting Notes Need A Clear Exit
A fleeting note should usually move toward one of:
- promoted insight into a permanent note
- archival as completed procedural history
- deletion if it was only scratch space
The vault should not treat all fleeting notes as equally durable by default.
Rule 5: File Artifacts Need Upstream References
If an artifact in 00_Raw/, 02_System/, 03_Web/, or 04_Experiments/ matters for later work, its existence should be pointed to by at least one of:
- a seam
- a handoff note
- a log entry
- a permanent or literature note
Draft Open Questions
- Should
handoffandfleetingremain separate types, or ishandoffbetter treated as a constrained subtype / pattern of fleeting work? - Should
log.mdbe treated as the canonical durable session summary, or should exported PoShWiKi session pages take that role more explicitly? - Should
agent-note-conventionsbe aligned downward to actual enforced minimums, or should the validator be aligned upward to the richer metadata model? - Should status values be explicitly enumerated in a single live source of truth?
- Should there be a formal rule for when to write both a seam and a handoff note versus only one?
Recommended Next Spec
I do not think the next step should be "write one big perfect protocol note."
I think the right sequence is:
- validate this artifact taxonomy
- choose one canonical source of truth for note
typeandstatus - define explicit decision rules for seam vs. handoff vs. log
- then write the final spec as a stable operational standard