Release Hikmah Stack v3.0.0 cognitive kernel

[CodeWithJuber]

What changed

Hikmah Stack v3 adds a deterministic Rust cognitive kernel beside the portable skill layer. The kernel is intentionally non-neural: models can propose language or hypotheses, but durable state, memory integrity, planning, provenance, contradiction handling, commitments, and verification remain inspectable and model-independent.

Cognitive mechanisms

• TraceWeave: typed immutable memory traces with dynamic context-sensitive recall
• Amanah Ledger: append-only BLAKE3 hash-chained memory/event ledger
• CounterTrace: explicit contradiction detection instead of silent overwrite
• Resonance Recall: multi-cue retrieval with salience, confidence, provenance, time, prospective urgency, and redundancy suppression
• Quiet Replay: consolidation proposals that preserve evidence and never silently rewrite truth
• Focus Capsule: bounded working set
• Promise Queue: prospective memory for commitments and deadlines
• Branch Loom: bounded symbolic planner over explicit world states/actions
• Parallel Deliberation Lanes: evidence, memory, risk, human impact, and delivery challenge lanes
• Model Port: architecture-agnostic proposal interface for local or hosted models
• Rust Truth Gate with a small Python compatibility fallback

Skills, playbooks, and lenses

Adds the cognitive-kernel skill and wires memory/verification behavior into Operator Core, Agent Radar, Decision Forge, Ship Guard, and Hikmah Orchestrator. Adds operational playbooks for memory lifecycle, parallel deliberation, and error-to-learning, plus lenses for memory integrity, model independence, and careful human-memory inspiration.

Safety and architecture boundaries

• No hidden chain-of-thought is persisted as memory.
• Sensitive traces are rejected by default unless policy explicitly permits them.
• Purge changes active state but does not pretend an append-only ledger has physically erased bytes.
• Neural/LLM systems are optional proposal engines, not the source of truth.
• Graph/vector databases are not mandatory architecture primitives.
• Human neuroscience is used as inspiration, not as a claim that this software reproduces the brain.

Validation

• Payload/source integrity verified before compilation
• cargo fmt --check
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test --workspace
• TraceWeave ledger/recall test passes
• Contradictory structured claims are not silently overwritten
• Branch Loom symbolic planning test passes
• cargo run -p hikmah-kernel -- validate --root .
• Six unique skills and v3 manifests validated
• Python compatibility hook compiles
• Clean production CI passed on commit aba52bdb6a9209849b9c6370345276f5764db2b1

This PR is intended to be squash-merged so temporary transport history does not enter main.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 49ce06a60d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Replace the corrupt bootstrap payload before marking it ready

When this commit is pushed to agent/cognitive-kernel-v3, adding READY triggers .github/workflows/bootstrap-hikmah-v3.yml, whose decode step concatenates every chunk and pipes the result through base64 -d | gzip -d. Reconstructing the payload at this commit produces SHA-256 c0c9e0b6..., not the declared 716adbd5..., and gzip -t reports CRC and length errors, so the workflow exits before applying the v3 patch. Regenerate the chunks and matching checksums before adding the trigger file.

Useful? React with 👍 / 👎.