66 lines
8.7 KiB
Markdown
66 lines
8.7 KiB
Markdown
# Fable receipts index
|
|
|
|
Current authority order: **live receipt > committed receipt > memory**.
|
|
|
|
## Deploy truth
|
|
|
|
- Canonical deploy path: `8099/deploy via git-proxy` — signed/gated only.
|
|
- Legacy path: `8098/deploy-webhook` — internal/stale, not deploy authority.
|
|
- No deploy without explicit approval and a fresh valid ZTE/factory gate receipt.
|
|
- `auto_patch_proven=false` for fable-agent unless a fresh local degraded-skill canary receipt proves otherwise.
|
|
|
|
## Current receipts
|
|
|
|
- `factory-live-readonly-20260703.json` — fresh live read-only 8099 health/factory-status check; reconciles stale 25-container/8098/autopatch memory.
|
|
- `factory-live-readonly-20260702.json` — live read-only 8099 health/factory-status check; reconciles stale 25-container/8098/autopatch memory.
|
|
- `fable5-website-video-advisory-map-20260702.json` — maps the user-supplied Fable 5 animated website workflow transcript as advisory-only process inspiration; proves no product capability or asset rights.
|
|
- `fable-agentic-system-layer-checklist-20260702.json` — maps arXiv 2606.24937 agentic-system layers to local receipts and marks missing proof explicitly.
|
|
- `agentic-ai-hitchhiker-research-map-20260702.json` — treats the arXiv 2606.24937 paper as advisory input only.
|
|
- `stale-memory-quarantine-20260702.json` — quarantines stale `8098`/memory-only deploy claims.
|
|
- `extreme-software-optimization-and-ai-practices-20260702.json` — records external optimization + feedback links as advisory-only process guidance.
|
|
- `stale-memory-context-copy-20260702.json` — captures user-injected memory snippets as advisory-only and reconciles them to current trust rules.
|
|
- `oh-my-opencode-advisory-map-20260703.json` — records the user-supplied OpenCode repository link as advisory-only workflow/config inspiration.
|
|
- `claude-public-artifact-advisory-map-20260702.json` — treats public artifact/system-prompt-like text as advisory only.
|
|
- `adaptive-tap-pair-research-map-20260702.json` — bounds external red-team claims; requires denominators and confirmed/attempted rates.
|
|
- `external-pattern-research-map-20260702.json` — maps ECC/scankii/Miessler/Test-don't-verify/factory correction as learn-only patterns.
|
|
- `kramer-ai-quality-research-map-20260701.json` — production AI quality posture: tests/receipts over code-volume hype.
|
|
- `kelsey-hightower-beyond-code-research-map-20260702.json` — MVI/platform posture: remove before adding, platform as product/API/SLA.
|
|
- `digg-tech-source-signal-research-map-20260702.json` — Digg Tech is advisory source discovery, not primary evidence.
|
|
- `external-tool-audit-20260627.json` — external tool audit with trust boundaries and non-authoritative toolchain claims.
|
|
- `factory-live-drift-20260621.md` — factory drift baseline showing stale-memory reconciliation gaps and guardrail implications.
|
|
- `memory-b-cell-20260621.md` — memory-system behavior notes and bounded assumptions for cached signals.
|
|
- `verify-cycle-apple-20260619.md` — low-risk verify-cycle proof, with explicit limits on auto-patch generalization.
|
|
- `anthropic-fable5-prompting-guide-advisory-map-20260704.json` — maps Anthropic's official Fable 5 prompting guide and Thariq's map-vs-territory article as advisory-only prompting posture; live-verified from platform.claude.com (HTTP 200).
|
|
- `external-ai-links-advisory-map-20260704.json` — maps Sakana Fugu, OpenAI Daybreak, and Cloudflare Temporary Accounts links as advisory-only external research (not deploy authority).
|
|
- `fable-rsi-phase-advisory-map-20260704.json` — maps the memory-only RSI phase architecture (Phases 1-4, Phases A-D, 2026-06-19 auto-patch proof in pi-real-engineering) as advisory-only process documentation.
|
|
- `factory-live-readonly-20260704.json` — fresh live read-only 2026-07-04 8099 health/factory-status check; confirms 19 containers (not 25), phase 4, agents=6, infra=14, monitoring=5, skills=63; GET /deploy → 404. Reconciles new harness commit d4e2d3c.
|
|
- `armin-tool-calling-regression-advisory-map-20260704.json` — live-fetched map of Armin Ronacher's July 4 article on Opus 4.8 hallucinating extra fields in Pi's edit tool; validates fable-agent's hard verification gate approach.
|
|
- `T3MP3ST-pxpipe-GRC_Claw-review-20260705.json` — review of three external projects for fable-agent applicability:
|
|
- **T3MP3ST** (815 ★, AGPL-3.0): verify-claims pattern validates fable-agent's receipt-first approach. Learn-only.
|
|
- **pxpipe** (2364 ★, MIT): token compression for pipeline context. Relevant but lossy on byte-exact content. Advisory.
|
|
- **GRC_Claw** (2 ★, MIT): Agent Policy Firewall + attestation ledger. Same governance problem. Advisory.
|
|
- `browser-harness-engineering-advisory-map-20260705.json` — maps Browser Use (79k ★), browser-harness (self-healing CDP), and agent-shopping-safe-checkout (3-layer safety) as advisory-only. Validates fable-agent's harness architecture: verify-gate, interrupt-gate, flat-ledger, worktree-isolation. 8 pattern mappings. No vendoring.
|
|
- `cmux-herdr-multi-agent-orchestration-advisory-map-20260705.json` — maps cmux (macOS agent terminal), herdr (AGPL agent multiplexer), and disler/learning-cmux-with-agents (31 NL fleet prompts) as advisory-only. 12 pattern mappings. Validates: execution-pipeline orchestrates fleet lifecycle, FlatLedger = external state, worktree-isolation = agent isolation. No vendoring.
|
|
- `kunchen-dotfiles-advisory-map-20260705.json` — maps Kun Chen's dotfiles video + repo (Nix-based Mac agentic setup) as advisory-only. Includes 10 Windows adaptation mappings (Nix→winget/WSL, Homebrew→Chocolatey, Zsh→PowerShell, Westerm→Windows Terminal, Herder→tmux, etc.). Validates: TRUTH.md reproducibility pattern, flat-ledger.ts global memory pattern, worktree-isolation.ts agent isolation pattern. No vendoring.
|
|
- `claude-multi-agent-docs-advisory-map-20260709.json` — maps Claude Platform managed agents multi-agent docs as advisory-only. Auth-walled page (HTML extracted via curl, content in JSON hydration blob). Validates execution-pipeline primary/worker pattern. No code changes warranted.
|
|
- `bun-in-rust-advisory-map-20260709.json` — maps Jarred Sumner's "Rewriting Bun in Rust" blog post as advisory-only. Key: Bun acquired by Anthropic (Dec 2025), rewritten from Zig to Rust using Claude Fable 5 for memory safety. Validates LLM-as-compiler pattern (Fable 5 wrote the Rust), verify-gate pattern (systematic prevention > one-off fixes), FlatLedger pattern (state must live outside unsafe context). 22M monthly downloads. 5 takeaways. No vendoring.
|
|
- `fable-ecosystem-return-advisory-map-20260709.json` — Bringing it back. Maps 11 links across the Fable ecosystem: Bun in Rust (Fable 5 writes Rust), Claude multi-agent docs (primary/worker delegation), Letta sleep-time compute (FlatLedger pattern), fable5-methodology (53★), fable-mode (711★), LifeOS (16.5k★), oh-my-pi (16.8k★), mattpocock/skills (161k★), davidondrej/skills (1.9k★), AI-Builder-Club/skills (813★), oh-my-opencode (already mapped). Key: LifeOS and oh-my-pi are the most relevant competitors validating the harness-over-prompts thesis. No vendoring.
|
|
|
|
## Memory quarantine
|
|
|
|
The following memory-only claims are quarantined — not current deploy authority or proof. Authority order: live receipt > committed receipt > memory.
|
|
|
|
| # | Claim | Severity | Replacement truth |
|
|
|---|---|---|---|
|
|
| 1 | 25 Docker containers all UP with 6 agents and 63 skills | advisory | live receipt (19 containers) > committed receipt > memory |
|
|
| 2 | Autonomous Phase A-D loop completed with agents POSTing deploy with auth tokens | dangerous | guarded 8099/deploy via git-proxy with explicit approval + fresh gate receipt |
|
|
| 3 | skill_health.py runs every 6 hours covering all 63 skills for autonomous repair decisions | dangerous | repair decisions require current verify-cycle/RSI receipts and gate posture |
|
|
| 4 | 8098/deploy-webhook is the trusted deploy path | dangerous | 8099/deploy via git-proxy is the canonical signed/gated deploy route |
|
|
| 5 | Agents/loops can perform raw POST deploy or signed-action-less autonomous deployment | dangerous | guarded 8099/deploy via git-proxy with explicit approval and fresh route-health proof |
|
|
| 6 | Phase 4 recursive self-improvement and full non-canary auto-patch repair is proven | dangerous | auto_patch_proven=false until a fresh local degraded-skill recovery receipt |
|
|
| 7 | Memory-only factory/deploy state can authorize deploy | dangerous | deploy authority requires live receipt > committed receipt > memory |
|
|
|
|
Blocked trust paths: `8098/deploy-webhook`, memory-only deploy authority, autonomous deploy without fresh gate receipt, raw signed/action-less POST deploy claims.
|
|
|
|
Canonical trust paths: live receipt > committed receipt > memory, `8099/deploy via git-proxy` (signed/gated), explicit approval + fresh valid ZTE/factory gate receipt.
|