feat: interop test package + session handoff doc

Cross-spec interop validation between ietf-act and ietf-ect:
- new packages/interop/ sibling package (ietf-act-ect-interop)
- 32 tests pass: shared claims, algorithm matrix, DAG structure,
  divergence handling, anti-goals
- documents ES256 raw signature wire-compatibility
- documents airtight typ separation (act+jwt vs exec+jwt)

Hazards surfaced:
- ACTLedger.append() silently accepts ECT Payload via duck-typing
  (both have .jti) — documented in interop README as a production
  hazard requiring external isinstance checks

Session handoff:
- SESSION-2026-04-12.md — snapshot of decisions, artifacts, open
  actions, and next-session starting points

Also: session-end commit of hash-format fix propagation to
packages/ect/ (the fix was applied to the old refimpl location
but did not propagate through the parallel package-move agent).
This commit is contained in:
2026-04-12 07:39:41 +02:00
parent 3a139dfc7e
commit 37859beef6
10 changed files with 1053 additions and 0 deletions

View File

@@ -0,0 +1,171 @@
# Session Handoff — 2026-04-12 (ACT / ECT IETF Strategy)
**Purpose**: Cold-start snapshot for the next session. Read this plus
`workspace/STRATEGY.md` to pick up without re-discovery.
---
## 1. Session Date & Context
- **Date**: 2026-04-12
- **Trigger**: Between 2026-04-07 and 2026-04-11, 14+ competing IETF
individual drafts and 7+ high-relevance arXiv papers appeared in the
agent-authorization / execution-accountability space. The window to
plant ACT + ECT as the standards-track home for this family is now
narrow. IETF 123 (July 2026) is the landing target.
- **Work mode**: Strategic consolidation — renames, restructure, diff
documents, outreach drafts, interop plan. No new normative content
added beyond what the existing drafts already carried.
---
## 2. Key Decisions Made This Session
1. **`par``pred` claim rename** (ACT). Aligns the ACT predecessor
claim with ECT's identical `pred` claim so both specs use the same
wire token for DAG parent references. Applied across draft text,
refimpl, tests.
2. **"Agent Compact Token" → "Agent Context Token" rename** (ACT).
Preserves the "ACT" acronym. "Context" better describes what the
token carries (invocation context — DAG refs, task metadata,
capabilities, delegation chain, oversight) and creates a clean
semantic pair with ECT (Execution Context Token).
3. **Package split into sibling packages `ietf-act` + `ietf-ect`** (no
shared core). Moved from monorepo with shared module to
`workspace/packages/act/` and `workspace/packages/ect/` as
independent packages. Decision: do not build a shared-core library
— the two specs have intentionally different scope, and forcing
shared abstractions obscures that. Cross-spec guarantees are
documented and tested via a separate `packages/interop/` plan.
4. **Position Option B chosen**: ECT is a WIMSE profile that
**normatively references** ACT. ACT is the general-purpose
primitive, ECT is the WIMSE-identity-bound execution profile. This
lets ECT be submitted for WIMSE WG adoption while ACT stays on the
independent-submission path (no WG dependency), and they cite each
other cleanly.
5. **`inp_hash` / `out_hash` format divergence acknowledged, not
unified this session**. ACT emits plain base64url; ECT validator
requires `sha-256:<b64url>` prefix. Documented as expected-xfail in
the interop test plan rather than forced into alignment — spec-level
decision deferred.
---
## 3. Artifacts Produced / Updated This Session
| Path (relative to `workspace/`) | Purpose |
|---|---|
| `STRATEGY.md` | Master strategy doc — landscape, positioning, phased action plan, risk register, success criteria |
| `packages/act/draft-nennemann-act-01.md` | ACT -01 draft; new §1.4.1 Related Work (concurrent proposals), new §1.5 Applicability (MCP/OpenAI/LangGraph/A2A/CrewAI/ECT), new §7.3 DAG vs Linear Delegation Chains |
| `packages/act/` (`ietf-act` refimpl) | 103 tests passing after `pred` and Context rename |
| `packages/ect/` (`ietf-ect` refimpl) | 56 tests passing; `inp_hash` bug fixed (removed stale `sha-256:` transformation in emitter path) |
| `packages/INTEROP-TEST-PLAN.md` | Planned `packages/interop/tests/test_interop.py` structure — shared-claim consistency, algorithm matrix, DAG cross-reference, claim divergence, anti-goals; user-facing compatibility matrix |
| `drafts/ietf-wimse-ect/draft-nennemann-wimse-ect.md` | ECT -02 (docname `draft-nennemann-wimse-ect-02`); normative ref to `I-D.nennemann-act` added |
| `drafts/ietf-wimse-ect/DIFF-vs-txn-tokens-for-agents.md` | ~1235-word factual diff doc vs `draft-oauth-transaction-tokens-for-agents-06` — claim-level matrix, lifecycle comparison, composition scenarios |
| `drafts/ietf-wimse-ect/wimse-intro-email.md` | ~390-word introduction email for wimse@ietf.org |
| `drafts/ietf-wimse-ect/ietf123-slides-outline.md` | 10-minute WIMSE slot outline: 10 slides, pacing plan, Mermaid diagrams for WIT/WPT/ECT layering and DAG-vs-linear, speaker notes, timing-discipline cuts |
| `drafts/outreach/emirdag-liaison-email.md` | Liaison email to Dr. Emirdag on SCITT-AI-agent-execution overlap; proposes cross-citation, claim alignment, possible joint IETF 123 slot |
| `drafts/outreach/oauth-ml-response.md` | Short oauth@ietf.org response to Txn-Tokens-for-Agents-06; frames ACT `pred` DAG as generalization of Raut et al.'s linear `actchain` |
---
## 4. Open Action Items (what the user does next)
Phase A items from `STRATEGY.md` still require execution:
- [ ] **A1**: Update ECT HTTP header section — replace `Wimse-Audience`
header with `wimse-aud` signature metadata parameter per
`draft-ietf-wimse-http-signature-03` (breaking change upstream,
published 2026-04-07).
- [ ] **A2**: Update SCITT refs in ACT to `draft-ietf-scitt-architecture-22`
(AUTH48); note "to be RFC-XXXX upon publication".
- [ ] **A3**: Lock Txn-Tokens refs in ACT/ECT to
`draft-ietf-oauth-transaction-tokens-08`.
- [ ] **A7**: Commit workspace + `research.ietf` subrepo changes.
Phase B outreach items (drafted but not sent):
- [ ] **B1**: Send Emirdag liaison email (`drafts/outreach/emirdag-liaison-email.md`).
- [ ] **B2**: Submit ACT -01 to datatracker.
- [ ] **B3**: Submit ECT -02 to datatracker.
- [ ] **B4**: Post ECT intro email to wimse@ietf.org, link DIFF doc.
- [ ] **B5**: Post OAuth ML response to oauth@ietf.org.
- [ ] **B6**: Request 10-min WIMSE slot at IETF 123.
- [ ] **B7**: Watch DAWN WG charter formation.
---
## 5. Pending Decisions (need user input)
- **Emirdag engagement depth**: liaison citation only, co-authorship
offer on a joint anchoring section, or just passive cross-citation?
The drafted email leaves all three doors open — pick one before
sending.
- **Refimpl publication to PyPI**: package names `ietf-act` and
`ietf-ect` are reserved but not published. User approval required
before any `twine upload`.
- **Repo strategy**: single monorepo for both drafts, or split into
separate Git repos so each draft has its own "home" for
kramdown-rfc / datatracker watchers? Current state is monorepo.
- **IETF 123 travel**: in person (Madrid) or remote? Affects
slide-prep cadence and whether to plan side meetings with Emirdag
/ Bertocci.
- **Hash encoding alignment** (ACT plain b64url vs ECT `sha-256:`
prefix): decide which spec moves, or keep divergence documented.
Interop plan currently pins it as xfail.
---
## 6. Known Landscape Threats (top 3)
1. **`draft-oauth-transaction-tokens-for-agents-06`** (Raut / Amazon,
2026-04-11). Linear `actchain` at OAuth AS layer — directly in the
same conceptual neighborhood as ACT. If this gets OAuth WG adoption
before ACT is visible, ACT has to position as "DAG generalization /
no-AS variant" instead of the default.
2. **`draft-emirdag-scitt-ai-agent-execution-00`** (VERIDIC, 2026-04-07).
AIR (AgentInteractionRecord) as SCITT payload. Not a direct
competitor but overlapping on input/output hashing, reasoning
capture, causality. Risk: if adopted first, ACT looks redundant
unless positioned as the *lifecycle* that AIR *anchors*.
3. **arXiv 2603.24775 (AIP / IBCTs)**. Closest *technical* competitor —
JWT + Biscuit/Datalog, exposes auth gap on ~2000 MCP servers, same
peer-to-peer-without-AS story. Not an IETF draft so no WG adoption
risk, but could become the citation of record in academic /
industry press if ACT is not visible fast.
(Full landscape table: `STRATEGY.md` §3.)
---
## 7. Next Session Starting Points (first 3 things)
1. **Read** `docs/control-center.md` (workspace root) and
`workspace/STRATEGY.md`; confirm Phase A is still the active phase.
2. **Execute A1** — patch ECT's HTTP header section to the
`wimse-aud` signature parameter form. This is the most urgent
technical fix; it's a breaking upstream change from
`draft-ietf-wimse-http-signature-03` and blocks ECT -02 submission.
3. **Execute A2 + A3** — refresh SCITT and Txn-Tokens reference
versions in both drafts so the submission snapshot is current.
After A1A3, move to Phase B (submissions + outreach sends).
---
## 8. Reference State Snapshot
- ACT refimpl: `packages/act/` — 103 tests pass, `pred` + Context
rename done, EdDSA + ES256 both supported.
- ECT refimpl: `packages/ect/` — 56 tests pass, `inp_hash` fix
applied, ES256 only, `exec+jwt` typ (legacy `wimse-exec+jwt` still
accepted).
- Interop package: not yet created; plan is in
`packages/INTEROP-TEST-PLAN.md`.
- Draft versions: ACT at `-01`, ECT at `-02` (`docname:
draft-nennemann-wimse-ect-02`).
- No submissions on datatracker yet this session (pending Phase A
completion).