Skip to content

AbstractSkill backlog — overview

Plain-language summary: abstractskill is the Agent Skills (SKILL.md) contract library for AbstractFramework (parse, validate, hash, discover, compose). The 2026-07-11 maintainer directive extends its role to SKILL TRUST: a system of validated skills, a do-not-use advisory registry, and (join-or-build) a network of trust for skills, with automated periodic reviews.

Counts

  • Planned: 5 (0001-0005)
  • Proposed: 2 (0006-0007)
  • Completed: 0
  • Deprecated: 0
  • Recurrent: 1 (advisory/validation registry review)
  1. 0003 simulated-execution audit harness — DESIGN done; implementation awaits runtime/agency lanes (upgrades the adopted shelf skills to audited, mints the first specific advisories).
  2. 0004 trust-network LEVERAGE step (backlog 0006) — consume external findings as advisory inputs with provenance.
  3. Consumer wiring — nothing consumes evaluate_trust yet (gateway console Phase 4 is the first consumer; contracts are ready).

Contract/registry/shelf/research/docs (0001/0002/0004/0005) shipped this session with 68 tests; 0003 is design-only by scope.

Planned items

ID Item State
0001 planned/trust/0001_trust_contract_layer.md Contract shipped (68 tests)
0002 planned/trust/0002_advisory_registry.md Mechanism shipped; registry empty (guidance carries class notices) until an audit/feed names a skill
0003 planned/trust/0003_simulated_execution_audit_harness.md Design done; execution awaits runtime/agency lanes
0004 planned/trust/0004_trust_network_research.md Position shipped (docs/trust-network-position.md)
0005 planned/trust/0005_first_party_shelf.md Shelf shipped (3 skills, hash-pinned)

Topic track: planned/trust/README.md (the skill-trust track, maintainer directive 2026-07-11).

Proposed items

ID Item Promotion criteria
0006 proposed/0006_automated_periodic_reviews.md Trust contract shipped + a scheduling host (gateway or CI) identified
0007 proposed/0007_admission_lint.md Phase-4 (entity-creation plan) opens; lint patterns validated against real advisories

Completed ledger

(empty — first entries land when 0001/0002/0005 close with evidence)

Deprecated

(none)

Recurrent

  • recurrent/advisory-registry-review.md — periodic re-verification of advisory/validation entries against upstream sources.

Process notes

  • Item IDs are four-digit and global across all lifecycle directories.
  • Completion appends a ## Completion report before the move to completed/.
  • Standing seat rules apply to every item: 1+ fable5 adversarial subagent per development, no commits without the maintainer's word, #FALLBACK-labeled degradations only.
  • 2026-07-11: backlog created (maintainer directive: validated-skills system, advisory registry, trust network, methodology skill, coredoc/backlog integration).
  • 2026-07-11: cycle 1 shipped 0001/0002/0005 core; cycle 2 folded a fable5 adversarial pass (class-advisory P0 → guidance split; graded severity; fail-closed verdict; adopted level; hash/format hardening) and wrote the 0004 trust-network position; cycle 3 applied the coredoc doc set (README, docs/*, architecture diagrams, llms.txt/llms-full.txt, SECURITY). 63 tests green. 0003 remains design-only (execution awaits runtime/agency lanes).