Draft safeguards

After generating a draft, WeaveRx runs fast, local-only checks — no LLM calls on the safeguard path. Flags are advisory; they never auto-block posting.

Check	Threshold (default)	What it detects
Entropy	Shannon > 5.5 bits/char	Possible encoded or obfuscated content
Length	> 6000 characters	Suspiciously long drafts
Repetition	ratio > 0.35 or repeated 4-grams	Heavy copy-paste or looped text
Patterns	regex heuristics	Credential-like strings, base64 blobs, private key markers, excessive markdown links (>15)
Relevance	keyword overlap < 0.08	Draft may be off-topic vs issue

Score	Status	Meaning
0.0 – 2.9	clean	No meaningful red flags
3.0 – 6.9	review_recommended	Skim draft before posting
7.0 – 10.0	high_risk	Multiple or severe heuristics fired

Disable: --no-safeguards or WEAVERX_SAFEGUARDS=0. Tune: WEAVERX_SAFEGUARD_ENTROPY_MAX, WEAVERX_SAFEGUARD_MAX_CHARS.

See the safeguard warning screenshot and sample_safeguard_warning.json.

Complements privacy keyword scanning. Safire is a separate path for deeper audit tooling.