Define Starintel capability-first boundaries #40

Merged

lost-rob0t merged 3 commits from agent/capability-first-cleanup into main

2026-08-06 01:23:42 +00:00

lost-rob0t commented

2026-07-24 16:25:31 +00:00

(Migrated from github.com)

Purpose

Define Starintel's capability-first mission without weakening runtime correctness.

What changed

broadens the canonical mission to collection, correlation, automation, and multi-source analysis
defines approval, audit, permission, retention, rate, review, redaction, and deployment governance as optional deployment modules
explicitly preserves mandatory parser, type, schema, poison-input termination, renderer sanitization/escaping, provenance, integrity, and replay invariants
updates mission alignment to prefer the strongest validated change consistent with user direction

Scope

AGENTS.md
skills/mission-alignment/SKILL.md

The stale July branch was archived as archive/pr40-capability-first-pre-rebase and rebased onto current main so no newer repository work is overwritten.

Validation

review feedback is addressed directly in the canonical agent contract
the mechanically edited backlog was removed from this PR rather than carrying optional correctness regressions forward
no runtime implementation code changed

## Purpose Define Starintel's capability-first mission without weakening runtime correctness. ## What changed - broadens the canonical mission to collection, correlation, automation, and multi-source analysis - defines approval, audit, permission, retention, rate, review, redaction, and deployment governance as optional deployment modules - explicitly preserves mandatory parser, type, schema, poison-input termination, renderer sanitization/escaping, provenance, integrity, and replay invariants - updates mission alignment to prefer the strongest validated change consistent with user direction ## Scope - `AGENTS.md` - `skills/mission-alignment/SKILL.md` The stale July branch was archived as `archive/pr40-capability-first-pre-rebase` and rebased onto current `main` so no newer repository work is overwritten. ## Validation - review feedback is addressed directly in the canonical agent contract - the mechanically edited backlog was removed from this PR rather than carrying optional correctness regressions forward - no runtime implementation code changed

lost-rob0t commented

2026-07-25 14:20:10 +00:00

(Migrated from github.com)

StarLang loop review

The capability-first direction is coherent, but this pass mechanically downgrades several runtime correctness invariants into optional deployment policy. These should be corrected before merge:

Deterministic validation — generated/LLM-produced artifacts must still be deterministically validated before they become typed StarLang documents or executable plans. Operator approval is optional; validation is not.
Schema validation — expert-system and actor outputs must remain schema-checked at the language boundary. Making this optional breaks typed actor contracts and the no-false-success rule.
Dead-letter handling — StarLang Loop 012 defines poison-input termination through reject/dead-letter behavior. The transport can expose pluggable dead-letter storage, but terminal handling itself cannot be optional under at-least-once delivery.
Renderer output safety — Content Security Policy may be a deployment module, but HTML sanitization/output escaping are renderer correctness requirements for untrusted document fields and generated views.
Document contract wording — lifecycle, provenance, integrity, and processing semantics needed for deterministic validation/replay should remain preserved. Optionality should apply to access/audit/retention/deployment-policy metadata, not to the evidence and execution model itself.

CI is not evidence yet: both Kindle EPUB and Org-roam Pages ended as action_required with zero jobs created.

Recommended correction: explicitly separate operator-selected policy modules from mandatory parser/type/schema/transport/rendering correctness invariants, while retaining the capability-first mission and direct-execution model.

## StarLang loop review The capability-first direction is coherent, but this pass mechanically downgrades several **runtime correctness invariants** into optional deployment policy. These should be corrected before merge: 1. **Deterministic validation** — generated/LLM-produced artifacts must still be deterministically validated before they become typed StarLang documents or executable plans. Operator approval is optional; validation is not. 2. **Schema validation** — expert-system and actor outputs must remain schema-checked at the language boundary. Making this optional breaks typed actor contracts and the no-false-success rule. 3. **Dead-letter handling** — StarLang Loop 012 defines poison-input termination through reject/dead-letter behavior. The transport can expose pluggable dead-letter storage, but terminal handling itself cannot be optional under at-least-once delivery. 4. **Renderer output safety** — Content Security Policy may be a deployment module, but HTML sanitization/output escaping are renderer correctness requirements for untrusted document fields and generated views. 5. **Document contract wording** — lifecycle, provenance, integrity, and processing semantics needed for deterministic validation/replay should remain preserved. Optionality should apply to access/audit/retention/deployment-policy metadata, not to the evidence and execution model itself. CI is not evidence yet: both `Kindle EPUB` and `Org-roam Pages` ended as `action_required` with zero jobs created. Recommended correction: explicitly separate **operator-selected policy modules** from **mandatory parser/type/schema/transport/rendering correctness invariants**, while retaining the capability-first mission and direct-execution model.