This release includes 3 breaking changes for platform teams planning a safe upgrade.
✓ No known CVEs patched in this version
Topics
+14 more
ReleasePort's take
Light signal`faf_go` now builds the canonical interview from faf-cli's public sources, and Turbo-Cat detection logic has been unified to use that same source list. The v5 guessing extractor has been removed.
Why it matters: Deprecation severity of 70 indicates a high‑impact change; teams relying on the v5 guessing extractor must update their metadata inference workflows before upgrading.
Summary
AI summaryRemoved the v5 guessing extractor and now compose interview and Turbo‑Cat detection directly from faf‑cli.
Changes in this release
| Type | Severity | Summary | CVE |
|---|---|---|---|
| Deprecation | High |
Removed the v5 guessing extractor; it no longer infers slots from project metadata. Removed the v5 guessing extractor; it no longer infers slots from project metadata. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Refactor | Low |
`faf_go` now composes the canonical interview from faf-cli's public sources, deleting local registry to prevent drift. `faf_go` now composes the canonical interview from faf-cli's public sources, deleting local registry to prevent drift. Source: llm_adapter@2026-06-12 Confidence: high |
— |
| Refactor | Low |
Turbo-Cat detection in `faf_formats` and `faf_auto` now composes faf-cli's single-source Turbo‑Cat list, removing the local 25‑entry copy. Turbo-Cat detection in `faf_formats` and `faf_auto` now composes faf-cli's single-source Turbo‑Cat list, removing the local 25‑entry copy. Source: llm_adapter@2026-06-12 Confidence: high |
— |
Full changelog
Every answer comes from one source. claude-faf-mcp now composes faf-cli's single-source engines — the 6Ws interview and Turbo-Cat detection — instead of carrying its own copies that drift. Fills come from real evidence or stay honestly empty; nothing is guessed. FAF don't lie, by construction.
Changed
faf_gocomposes the canonical interview — sourced from faf-cli's publicINTERVIEW/interviewForMissing; the local question registry is deleted, so it can't drift from the kernel. Only missing + active slots are asked (slotignoredskipped).- Turbo-Cat detection composes faf-cli —
faf_formats/faf_autouse faf-cli's single-source Turbo-Cat (~200 formats); the local 25-entry copy is deleted. Correct detection that can't drift (an mcpbmanifest.jsonis no longer mis-flagged as a Chrome extension). /fafdrives to a verified 100% — runsfaf_trust(✪ parity receipt), offersfaf_setupto keep it fresh, fills only from sourced evidence.
Removed
- The v5 guessing extractor — inferred slots from project name / tech stack. Gone; replaced by a sourced-or-empty extractor with a test proving it can't fabricate. No-guess is now structural.
35 tools. 567 tests, 0 fail. Composes faf-cli ^6.10.1.
FAF defines. MD instructs. AI codes.
Breaking Changes
- Removed the v5 guessing extractor that inferred slots from project name/tech stack.
- Composed `faf_go` interview directly from faf-cli's public `INTERVIEW` / `interviewForMissing`; local question registry deleted.
- Composed Turbo‑Cat detection directly from faf-cli (~200 formats); local 25‑entry copy removed.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About Wolfe-Jam/claude-faf-mcp
First & only persistent project context MCP. Provides .faf (Foundational AI-context Format) Project DNA with 33+ tools, Podium scoring (0-100%), and format-driven architecture. Official Anthropic Registry. 10k+ npm downloads.
Related context
Related tools
Earlier breaking changes
- v5.9.1 Registry namespace changes to `one.faf/*`
Beta — feedback welcome: [email protected]