artale
/
fable-agent
Template
1
1
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

chore: quarantine A-D autonomous deploy memory and index anthropic prompting guide

This commit is contained in:
artale 2026-07-04 15:33:29 +02:00
parent 78667dd93c
commit 90b36c9e22
2 changed files with 55 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
docs/receipts/anthropic-fable5-prompting-guide-advisory-map-20260704.json Normal file
View File

@ -0,0 +1,49 @@
{
"schema": "fable.research_map.receipt.v1",
"createdAt": "2026-07-04T00:00:00.000Z",
"taskLabel": "anthropic-fable5-prompting-guide-advisory-map",
"task": "Map the user-supplied Anthropic Fable 5 prompting guide and Thariq map-vs-territory framing as advisory-only prompting posture",
"source": {
"classification": "user-curated external link with framing",
"items": [
"https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5",
"https://x.com/trq212/status/2073100352921215386"
],
"contentVerified": false,
"localProofExists": false
},
"extractedPrinciples": [
"set effort to high by default, xhigh for hardest work",
"give it the goal and the reason, not step-by-step commands",
"add a memory file so it learns from past runs",
"make it verify progress against real tool results",
"use fresh verifier subagents, they beat self-critique",
"stop telling it to explain its reasoning, that triggers refusals",
"map vs territory: quality is bottlenecked by how well you uncover unknowns before and during implementation",
"better models don't remove the need to think; they make unclear thinking more expensive",
"every brainstorm, prototype, interview, reference, implementation plan and explainer is a cheap way to discover what you didn't know before it becomes expensive to fix"
],
"constraints": [
"external link is advisory only until locally verified",
"these principles are prompting guidance, not local deployment authority",
"canonical deploy remains 8099/deploy via git-proxy",
"8098/deploy-webhook remains legacy/internal/stale for fable-agent trust",
"auto_patch_proven remains false without a fresh local degraded-skill recovery receipt",
"no deploy, vendoring, dashboards, or new packages were performed for this mapping"
],
"fableAgentAlignment": [
"fable-agent already uses goal-with-reason prompts (ADW pipeline, verify-cycle, receipts)",
"memory file pattern already exists via memory b-cell and stale-memory quarantine receipts",
"verifier subagents are used by verify-cycle and external-project-review receipts",
"fable-agent does not request reasoning explanations in production loops (receipts prove, not reasoning dump)",
"map v territory framing validates the receipt-first methodology: receipts are the territory, memory is the map"
],
"skipped": [
"live fetch/content verification from the provided links",
"deploy actions or route checks",
"new dependency changes"
],
"decision": "advisory",
"deployAttempted": false,
"autoPatchProven": false
}

6
docs/receipts/stale-memory-quarantine-20260702.json
View File

@ -9,6 +9,12 @@
"replacementTruth": "container/deployment authority follows live receipt > committed receipt > memory; reconcile against a fresh live factory-status/checker receipt",
"severity": "advisory"
},
{
"claim": "memory-only autonomous Phase A-D loop completion with agents posting HTTP POST deploy commands using auth tokens",
"staleBecause": "repeated live and committed receipts confirm auto_patch_proven=false; any claim that agents can autonomously POST deploy with auth tokens is memory-only",
"replacementTruth": "agents must use guarded 8099/deploy via git-proxy with explicit approval and fresh ZTE/factory gate receipt; raw command/AWS-like POST deploy is blocked",
"severity": "dangerous"
},
{
"claim": "skill_health.py runs every 6 hours and covers all 63 skills for autonomous repair decisions",
"staleBecause": "cron and full-skill coverage are memory claims without an in-session live proof for this chain",