Nick
b0c287ba8b
feat(runtime): batch — plan loop 4.2, self-transformation 6.2, taxonomy closure 6.3, taxonomy dispatch fix
...
- 4.2 Plan loop runtime (tui/planloop): full consent-gated process —
contextual offer (sustained-pattern or explicit), never-interrupts-
active-Task, initial plan -> research -> viability -> revised ->
targeted research -> final viability -> owner-gated start ->
dependency-ordered execution -> independent review -> report;
return-to-plan on contradiction (round++, targeted re-research,
start gate re-arms, done steps survive); durable/resumable/piecewise
through the K25 plan store. 8 frozen tests red-first x5 green.
- 6.2 Self-transformation pipeline (src/rune/self_transform_pipeline.py):
13 stages, PROPOSE->...->ROLLBACK_OR_CLOSE; DEP06 mediation at stage 4
+ shadow wrap (audit catches hidden writes); DEP07 activation at stage
6; DEP05 owner envelopes at stage 8 (authenticated, reserved, then
REFUSED — SELF_ACTIVATION_IMPOSSIBLE per D10, stages 9-11 structurally
unreachable); D03 render, D05 per-effect, D06 vintage, D08 rollback-
pause; journal receipts digest-verified. 36/36 frozen tests red-first.
- 6.3 Taxonomy closure: NO taxonomy code ships — scan conformance
(planted router tripped RED, removed, 8/8 green); authority path
unchanged (user: envelope still required, 53 passed); TUI unblocked.
8 frozen tests red-first. Taxonomy dispatch marker fix.
2026-10-17 22:17:31 +01:00
Nick
834eece4d7
feat(runtime+facade): batch landing — receipt presentation, compaction, composer depth, themes, Sleight wiring, gitignore hygiene, taxonomy
...
- K27 receipt presentation: full decision+evidence+effect packages,
expandable sections, UNKNOWN-never-success in every dimension state,
spine verdicts via evidence.Evaluate (read-only import)
- 2.6-depth compaction: append-per-message (O(1) tail line, 198 bytes
steady-state), compaction every 64 appends (512/channel hot), cool
archives per memory-continuity contract, nothing deleted, exact-by-ID
across compaction+restart+tamper, legacy checkpoints heal on load
- K27 composer depth: 3-line input per old TUI contracts (§2.3 history
walk, placeholder, stash/restore), deterministic natural-language
resolution (40+ pinned phrases), palette natural-tier + ctrl+p bind,
Accessibility profile (keyboard-only/no-color/reduced-motion)
- K27 themes: 10 fixed Go theme tables (byte-exact from Textual 8.2.8,
not guessed; nord/dracula-default/monokai/solarized/catppuccin/tokyo-
night/flexoki/gruvbox), /theme command (mirrors _cmd_theme exactly),
live-editor data shape (staged per-token preview overrides), base bg
rgb(30,30,40), no-drift drift test anchoring Go to retiring Python
- Sleight live wiring: operator._context_broker registers seventh
contributor 'continuity' (bucket 2, fail-closed); K16 allowlist
amended (memory_continuity.working_set); matrices now run against
the live road; 102 tests green
- .gitignore hygiene: directory-level traps fixed (internal/ re-includes
.go/.json/.md only; root-anchored /index/ /journal/ /state/; fixtures
+ testdata re-included at any depth); 6-test frozen gate; unified_check
member wired; real git add (no -f) proven on testmod/internal
- Receipts, compaction, composer, themes, sleight-wiring, hygiene all
under ADMIT contracts (101+ contracts, 0 research-gate violations)
- main.go NOT landed (chat-live P1 wave owns it, still in flight)
2026-10-17 22:00:03 +01:00
Nick
122099faa8
feat(openspec): D31 — TUI complete-on-arrival; GUI offers both response display modes (owner directive: TUI is speed and cleanliness)
2026-10-17 18:32:03 +01:00
Nick
34e78f2e53
feat(runtime): Sleight matrices 3.4b/c/d — research-first (PROCEED), 39 tests, 13 fault receipts, §18 deliverables
...
Research chain research/factory-sleight/ (full protocol, QUALIFY PROCEED —
model-dependent metrics honestly DEFERRED to B-CHAT-LIVE, never faked).
3.4b: S01-S15 scenario rows, one deterministic each against the exact
staged generation. 3.4c: §16.1 invariants 100% + measured semantic
metrics (recall integrity 1.0, topic resume 1.0, commitment loss 0.0,
pointer liveness 1.0) + 13 fault-injection fixtures (crash/read-only/
disk-full/network/embedding-outage/stale-index/duplicate-replay/forced-
budget/child-failure) each with degraded/recovery receipts, all
recoverable. 3.4d: §18 deliverables — ownership diagram, module audit,
UI states, schemas, phased plan, guarantee boundary. 69/69 green incl.
K15+K16 boundary intact. Main-session verify: green own run.
2026-10-17 17:49:47 +01:00
Nick
b6ff9df9e7
live-proof probe
2026-10-17 11:07:31 +01:00
Nick
d260b6a4af
feat(openspec): D30 — WASM-standard boundary law (owner directive)
...
WASM via wazero = the standard Rust crossing (speed class + real
isolation + designed debug errors + no cross-compile pain); cgo
exception-only with named re-provable conditions; five libraries
in-process; crossings batched; prototype round gates the migration.
2026-10-17 06:36:26 +01:00
Nick
c5b4a61cc9
feat(openspec): D29 + TD-4 — selection law: click/shift+tab, highlight = input target (owner ruling)
2026-10-16 20:25:10 +01:00
Nick
d42b0b31e6
feat(openspec): D28 + TD-1..3 — question-anchored fractal task trees (owner ruling)
...
Tree spawns from a question, grows to resolution, then a new tree; threads
nest as trees-within-trees; always visible, fully live, fully workable
(A+C blend). Binding on task-lane live wiring.
2026-10-16 19:58:56 +01:00
Nick
fd6ddc4d89
feat(openspec): D27 + phase 0D — versioned releases + in-TUI update offers (owner directive)
...
Tag-driven semver injection through the paired-build stamp; tui-v* tag
pushes cut Gitea releases with the vendored binary attached; update
offers go through the self-transformation + migration rules (offer never
silent, consent + digest verify + restore point + honest post-restart
version; owner's Gitea only). Tasks 0.16-0.18. Strict-valid.
2026-10-16 11:20:16 +01:00
Nick
74a8ffd27d
feat(openspec): D26 + phase 0C — unified build & test substrate (owner directive)
...
Owner ruling 2026-10-15: no two test worlds — the system is tested as one.
D26 makes the dichotomy's known failure modes structurally impossible:
generated boundary shapes (one schema source, both sides code-generated,
hand-written shared shapes gate-rejected), one unified runner (Rust + Go +
boundary round-trip corpus + verdict fixtures + mutants; partial green
impossible), paired builds with startup attestation (mismatch fails
closed; no committed compiled artifacts), and the landing rule (every
landing passes the unified runner). Phase 0C tasks 0.12-0.15 added before
construction begins. Strict-valid.
2026-10-15 22:40:24 +01:00
Nick
b5ab45188d
feat(openspec): SA-D1..D7 amendment — sub-agent orchestration, living task tree, thread model + tree view, day-one sockets
...
- NEW capabilities: sub-agent-orchestration (full live view/control, standing
rule all the way down, per-action receipts, chained spawn with ancestry),
task-management (living task tree: statuses, deps, blocked-by, reorder,
Rune/agent assignment, history), thread-orchestration (universal thread
unit, origin-to-root traceability, start/split with marked divergence,
agent workspaces in one tree)
- future-seam-registry: day-one sockets requirement (agents, tasks, threads,
messaging, cognition-reserved, connectors; name + contract + deletion test)
- visual-system: terminal-native thread tree view + linear fallback
- dual-lane + alerts: agent/task-originated messages follow existing rules
- proposal (19 capabilities), design (D21-D25), tasks (2.8-2.10, sockets gate)
- openspec validate --strict: VALID — 135 requirements, 216 scenarios
Owner decisions SA-D1..D7 recorded in research/harness-survey/interviews.md.
Altitude frozen per owner: behavioral spec complete; finer detail deferred to
Task 0.0 gates, visual round (post semantic map), and future rounds.
2026-10-15 18:06:37 +01:00
Nick
65fef8cd3b
feat(openspec): add-native-rune-tui complete spec packet — 16 capabilities, R2 frontier expansion, strict-valid
...
- proposal/design/tasks rewritten: R2 frontier scope folded in per owner
interview R2-D1..R9 (2026-10-15); banner removed; 16 capabilities
- 14 capability deltas expanded (trust envelope w/ no-dead-end amendment,
web-link live feed replacing multi-face TUI, real-hardware qualification
gates, completion-receipt evidence spine, attestation + mutant gates)
- 2 NEW capabilities: extension-lifecycle, future-seam-registry
- accessibility parity claim removed (R2-D1 deferral); visual selection
sequenced after semantic map (R2-D7); calendar-first connector (R2-D8)
- phase 0B pre-build repairs: five live defects as activation prerequisites
- research evidence chain included: harness-survey (interview ledger,
18 interviews closed), tui-frontier-r2 (8-stage pipeline, QUALIFY:
PROCEED), voice-research (speech pipeline, DEFER_AND_DESCOPE, four
reserved seams)
- openspec validate add-native-rune-tui --strict: VALID
Refs #2484 (epic), closes nothing yet — review approved by owner;
TUI build phase unblocked on merge of this spec baseline.
2026-10-15 17:06:45 +01:00
Nick