agentleFS
stevehansenstevehansen/eidet/.claude/skills/writepath/SKILL.md

writepath

Skill4 starsChanged 54 days agoMITView on GitHub

Prime on Eidet's write path and integrity model before changing it — the always-on secret scanner, the signal/self-talk gate, WriteValidator entry construction, duplicate and conflict gates, quarantine, the poison log, derived MemoryTrust, provenance tiers, content commitments, and the IntegrityAuditor. Use when the task touches WriteValidator, SecretScanRule, ConflictGate, MemoryTrust, MemoryCommitment, ProvenanceRules, QuarantineInfo, IPoisonLog, IntegrityCheck, or any STRIDE tampering/poisoning finding. Not for the MemoryEntry shape or lifecycle verbs (see memory), not for recall ranking (see recall), not for scheduled rewrites (see maintenance).

The file

Files
SKILL.md

Write path & integrity — priming

Canonical spec: docs/domains/writepath.md — read it for the full gate order, all invariants, key files, and gotchas. Terms of record: UBIQUITOUS_LANGUAGE.md § Write path. Threats: STRIDE.md T-8, T-13, T-15, T-20.

Deterministic, local, zero-LLM. The posture is adversarial: assume some memories are poisoned, imported, or rewritten, and make sure they cannot pass as first-party knowledge. memory owns the entity; this domain owns whether a write happens and what trust it earns.

Core invariants (get these right)

  • Secret scanning is never optional on any write path. ScanSecrets is separate from Validate so non-semantic surfaces can skip the low-signal/self-talk gates — never the secret scan.
  • WriteValidator.BuildEntry is the only construction path for a stored memory (validation, id, defaults, provenance, entity extraction). Don't hand-build a MemoryEntry on a write path.
  • Trust is derived on every read, never stored — no field to forge or let drift.
  • The commitment factor multiplies AFTER the echo lift. Reversed, echoes launder rewritten content (STRIDE T-8). The only sanctioned repair for a broken commitment is supersession.
  • No insecure default in ProvenanceTrust — trusted origins are enumerated, everything else (including Unknown) lands on the import floor. Unknown sits at exactly that floor because the pack-import clamp compares floors.
  • Three tiers: 1.0 first-party, 0.7 intake, 0.5 import. Intake reads local repo files, so it is not the remote-import surface — holding it at 0.5 made repos' own docs unreachable behind session chatter (measured: hit@5 53% vs 94%). A pack declaring provenance=intake is clamped to Pack, since the intake tier now outranks it.
  • A synthesis inherits its least-trusted contributor unless every contributor is trusted (ProvenanceRules) — that's the anti-laundering guarantee.
  • De-boost, never hide. A quarantined or broken-commitment memory stays recallable so it can earn the echo that clears it; it raises a dashboard finding instead of disappearing.
  • A contradiction needs all three: near-duplicate content + opposite hard valence signs + high-trust incumbent. Supersessions are exempt from the conflict gate and the poison fast-path.
  • Adding an IntegrityCheck without a probe throws — and that throw is exempt from per-check isolation, on purpose.

Key files / reuse

  • src/Eidet.Core/Gates/WriteValidator.cs + SecretScanRule.cs — the gate chain.
  • src/Eidet.Core/Memory/MemoryTrust.cs — the trust algebra (Explain for forensics).
  • src/Eidet.Core/Memory/MemoryCommitment.cs — Render is the only way to authorize an in-place content rewrite; anything else reads as tampering.
  • src/Eidet.Core/Memory/{ConflictGate,ProvenanceRules,IPoisonLog}.cs — write-time verdicts.
  • src/Eidet.Core/Integrity/IntegrityAuditor.cs — one probe per read path.

Gotchas

  • The secret gate is pattern-based, not entropy-based: prose mentioning Password=… or a base64 blob ending == is rejected. Rephrase the memory; don't weaken the pattern.
  • Stores reject secrets; only the memory-tool path redacts them.
  • BuildEditEntry stamps Provenance = UserStated unconditionally — an edited agent memory becomes fully trusted.
  • NullPoisonLog is the default, so the poison fast-path is inert unless the host wires the Raven log.
  • A clean IntegrityReport is only as broad as ChecksProbed; capped citations and failed probes are unprobed, not clean.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.