Release tracker: 0.5.0 — Intelligence + Integrity

[prakashUXtech]

Status — 2026-05-13

Captain track shipped. All 9 bundled PRs landed in dev:

• Trust hardening: feat(trust): timestamp monotonicity + canonical JSON hardening + payload typing + key rotation tests #210 ✓ · feat(trust): audit log payload summaries + structured chain-append-failure logging (#201, #202) #213 ✓ · feat(trust): touch-time chain pruning with configurable cap (#203) #214 ✓ (touch-time stub; full archival deferred to 0.5.x)
• Memory primitives: feat(memory): brain-aligned memory update primitives (#192) #224 ✓ (RFC docs(rfc): memory update primitives — confirm/update/supersede with prediction-error gating (#192) #209 ✓, spike + prod folded into feat(memory): brain-aligned memory update primitives (#192) #224)
• Intelligence: feat(eval): YAML eval format and runner for memory-driven agents (#160) #211 ✓ · feat(memory): graph traversal + progressive loading + typed entity ontology (#108, #190) #219 ✓ (covers Graph memory traversal and progressive loading #108 + feat(graph): tier overhaul — typed ontology, real extraction, graph API #190) · feat(optimize): autonomous self-improvement loop using eval signal (#142) #223 ✓
• CLI: feat(cli): soul journal shell-hook subcommand (#189) #212 ✓ · feat(cli): soul diff structured comparison (#191) #215 ✓

Plus #233 (soul note + observe), #221 (graph plateau fix for #220).

LFDT mentee track in flight:

• chore(cli): deprecate soul remember in favor of soul note (phase 2 #231) #236 AnshulPatil2005 — deprecate soul remember — APPROVED, awaiting CHANGELOG + horizon
• feat(test): Adding cross runtime tests suite for PocketPaw, LangChain, CrewAI #232 Akshat-Raj — cross-runtime tests — CHANGES_REQUESTED (retarget to dev, real PocketPaw subprocess test)
• feat(trust): adding digital signature verification on import #237 Akshat-Raj — trust import verification — CHANGES_REQUESTED (retarget to dev, legacy-soul WARNED semantics, tuple-shim path)

Remaining for v0.5.0 release:

• Fix bug bug(memory): episodic writes wrap plain content in User:/Agent: envelope #234 (episodic User:/Agent: envelope wrap)
• Land mentee PRs above
• Optional hardening: trust: harden _canonical_json against non-JSON-native types #200 · trust: tests for key rotation (mixed-signer chains) #204 · trust: tighten compute_payload_hash typing + reject Pydantic models #205 (good-first-issue trust polish)
• Release PR dev → main

Auto-closed via this update: #108, #142, #160, #189, #190, #191, #192, #199, #201, #202, #220.

---

(Original tracker body below for reference.)

Tracker for the v0.5.0 minor release. Major bump because the schema grows (#192 brain-aligned memory primitives) and several new top-level subsystems land (soul-aware evals, soul optimize autoresearch, graph traversal API).

Theme

Two pillars:

1. Intelligence — souls evaluate themselves (Soul-aware evals: a format for testing memory-driven agents #160), improve themselves (feat: soul optimize — autonomous self-improvement loop (autoresearch) #142), and traverse their own graph memory (Graph memory traversal and progressive loading #108 + feat(graph): tier overhaul — typed ontology, real extraction, graph API #190).
2. Integrity — better memory updates (Memory updates: brain-aligned primitive set (supersede + confirm + update + weight-decay forget) #192), tighter trust chain (trust: enforce timestamp monotonicity in verify_chain #199-trust: tighten compute_payload_hash typing + reject Pydantic models #205), audit-log polish (trust: payload summaries in audit_log for human-readable history #201, trust: WARNING-level logging when chain append fails for non-readonly reasons #202), chain pruning (trust: chain pruning + archival for long-running souls #203 stub), key rotation foundation (trust: tests for key rotation (mixed-signer chains) #204).

Plus quality-of-life CLI surfaces (#189 `soul journal`, #191 `soul diff`).

Bundled work — 14 items, 9 PRs

Trust chain hardening + observability

Issue	PR	Branch	Status
#199 + #200 + #205 + #204	#210	feat/trust-chain-verification-hardening	Open against dev
#201 + #202	#213	feat/audit-log-and-logging-improvements	Open against dev
#203 (stub)	#214	feat/touchtime-chain-pruning	Open against dev

Memory primitives (#192)

Step	Branch	Status
RFC	#209 (rfc/memory-update-primitives)	Open against dev — captain reviewing
Spike	feat/memory-update-primitives-spike	Pending — waits on RFC sign-off
Production	feat/memory-update-primitives	Pending — waits on captain daily-test of spike

Intelligence

Issue	PR	Branch	Status
#160	#211	feat/soul-aware-evals	Open against dev
#108 + #190	TBD	feat/graph-memory-traversal	In flight
#142	TBD	feat/soul-optimize-autoresearch	Bench — waits on #211 merge

CLI polish

Issue	PR	Branch	Status
#189	#212	feat/soul-journal-cli	Open against dev
#191	#215	feat/soul-diff-cli (stacked on #212)	Open against dev

Schema migrations

`soul migrate ` covers 0.4.x → 0.5.0:

• `MemoryEntry.retrieval_weight: float = 1.0` (new in Memory updates: brain-aligned primitive set (supersede + confirm + update + weight-decay forget) #192) — backfilled from `importance` for legacy entries
• `MemoryEntry.supersedes: str | None = None` (new in Memory updates: brain-aligned primitive set (supersede + confirm + update + weight-decay forget) #192) — derived from existing `superseded_by` reverse-mapping
• `MemoryEntry.prediction_error: float | None = None` (new in Memory updates: brain-aligned primitive set (supersede + confirm + update + weight-decay forget) #192) — None for legacy
• Graph layer (Graph memory traversal and progressive loading #108 + feat(graph): tier overhaul — typed ontology, real extraction, graph API #190) — typed entity ontology, requires re-extraction. Auto on awaken.
• Trust chain — additive (the chain.pruned action and TrustEntry.summary field are forward-compatible)
• Keystore — `previous_public_keys` allow-list (new in feat(trust): timestamp monotonicity + canonical JSON hardening + payload typing + key rotation tests #210)

Workflow

• Sub-PRs target `dev`, not main directly
• Captain installs dev locally and tests through daily soul use
• Once dev is stable, dev → main via release PR with full test matrix
• Tag `v0.5.0` on main, PyPI publish workflow triggers

Out of scope (deferred to v0.5.x)

• Full trust-chain archival (`trust_chain/archive/` directory + `chain.archived` checkpoint entries) — only the touch-time pruning stub ships in v0.5.0
• `soul rotate-keys` ergonomic CLI — rotation is API-driven in v0.5.0; CLI wrapper later
• Cross-soul knowledge graph federation
• LLM-driven entity disambiguation
• Diff tool enhancements (soul diff enhancements (post-#191) #217)
• Soul public API cleanup (Soul-side public API cleanup — diff tool surface findings #216)

Estimated agent-time

~46-63 agent-hours across 9 PRs. Wall-clock with 5 parallel agents: 3-5 days of crew work + 2-5 days of captain daily-testing the dev branch + 1 day for the final release PR.

Closes (when v0.5.0 ships)

#199, #200, #201, #202, #203 (stub), #204, #205, #189, #191, #160, #108, #190, #142, #192. Umbrella tracker #206 (0.4.1) is fully absorbed into this v0.5.0.