pdurlej/platform

Fork 0

WIP: feat(agent-access): v0 4-slice skeleton — session-scoped credential delivery [#76] #323

Closed

claude wants to merge 1 commit from claude/feat-agent-access-plane into main

claude commented

2026-05-17 10:17:51 +02:00

Collaborator

Summary

Pre-impl skeleton + Spec Kit for issue #76 — Agent Access Plane.

Status: draft — for review, NOT for merge.

Pattern: hybrid GitHub Spec Kit + code skeleton, same as pdurlej/iskra-openclaw PR #276 / #278 / #279 / #281.

Author: claude. Implementer: codex (anticipated). Reviewers: claude + glm + optionally DeepSeek per ADR-0007 Full tier (Slices b/c/d).

Source: #76 (codex-authored, parent for #56, #73, #74).

What's in here

13 files, ~1300 LOC. Three categories.

Spec Kit markdown (6 files, ~1000 lines)

docs/specs/agent-access-plane-v0/:

00-constitution.md — 8 principles (no standing credentials, Infisical as source of truth, ssh-agent with TTL, per-cousin machine identities, wrapper-injected secrets, runtime state non-secret, capability catalog enforcement, per-session operator approval), 10 forbidden anti-patterns, platform vs iskra-openclaw boundary, North Star.
01-specify.md — 11 FRs, 6 NFRs, 5 open questions, 8 falsifiable success criteria, explicit out-of-scope.
02-plan.md — 4 slices (a/b/c/d), M1-M5 milestones, 6 risks + mitigations, rollback strategy per slice, composability proof.
03-tasks.md — atomic task list, ~2200 LOC v0 estimate, codex execution notes (one slice = one PR per ADR-0017, adversarial tests mandatory).
04-implement-notes.md — 13 pre-impl decisions (D1-D13: YAML over JSON, ULID over UUID, JSONL over SQLite, lazy TTL, out-of-band approval, operator-only critical tier, composability over completeness, single binary, per-session ssh-agent, wrapper-script for command capabilities, cousin-PAT for low-tier, runtime state non-sacred, ...), cross-reference matrix, known unknowns, codex pickup hints.
README.md — index + per-cousin reader guide.

Code skeleton (py_compile clean)

control-plane/platformctl/agent_access/:

_types.py — Capability / Session / AuditEntry / SessionStatus / AuditLevel / BackingType / CapabilityBacking dataclasses
catalog.py — load_catalog, validate_catalog, lookup_capability stubs with TODO(codex Slice a) markers pointing to specific 03-tasks.md entries

Schema + seed catalog

schema/agent-capability.schema.json — JSON Schema draft 2020-12, validates schema_version + capabilities array with type-discriminated backing (ssh-agent / wrapped-command / infisical-token / forgejo-pat).
policies/agent-capabilities.yaml — 5 seed capabilities covering low/medium/high/critical audit tiers:
- forgejo-pat-read (low, all reviewer cousins)
- ssh-rs2000-platform-host-agent (high, codex-only, deploy operations)
- ssh-vps1000-iskra-readonly (high, codex+claude, inspection only)
- infisical-secrets-read-scoped (medium, scoped path)
- break-glass-full-access (critical, operator-only via grant)

Contract tests

control-plane/platformctl/tests/agent_access/test_catalog.py — 10 test scaffolds (xfail strict=False) covering load/validate/lookup interface contract. Codex flips xfail off when implementing Slice (a).

Cross-references

ADR-0003 Agent Access Plane (parent) — to be reconciled at M5
ADR-0005 coordination lanes, ADR-0010 cousin role taxonomy, ADR-0017 no stacked PRs (one slice = one PR), ADR-0018 fix root causes not workarounds (no "accept workaround" outcomes)
Composes with: #56, #73, #74, #132, #198, #199, #200
Supersedes: pdurlej/iskra-openclaw#43 canonical credential architecture

Tier

This PR: Trivial (draft scaffold, no runtime/schema/sacred-path mutation, py_compile clean stubs).

Future slice PRs: Lite (Slice a, catalog schema), Full (Slices b/c/d — security-sensitive, sacred-path-adjacent).

Operator action

Review scope. If accept → mark this PR draft-cleared OR close + signal codex to start Slice (a) in a fresh PR (per ADR-0017 no-stacked-prs, the Spec Kit + impl are separate PRs).

Refs #76

## Summary Pre-impl skeleton + Spec Kit for **issue #76 — Agent Access Plane**. **Status**: `draft` — for review, NOT for merge. **Pattern**: hybrid GitHub Spec Kit + code skeleton, same as pdurlej/iskra-openclaw PR #276 / #278 / #279 / #281. **Author**: claude. **Implementer**: codex (anticipated). **Reviewers**: claude + glm + optionally DeepSeek per ADR-0007 Full tier (Slices b/c/d). Source: [#76](https://git.pdurlej.com/pdurlej/platform/issues/76) (codex-authored, parent for #56, #73, #74). --- ## What's in here 13 files, ~1300 LOC. Three categories. ### Spec Kit markdown (6 files, ~1000 lines) `docs/specs/agent-access-plane-v0/`: - `00-constitution.md` — 8 principles (no standing credentials, Infisical as source of truth, ssh-agent with TTL, per-cousin machine identities, wrapper-injected secrets, runtime state non-secret, capability catalog enforcement, per-session operator approval), 10 forbidden anti-patterns, platform vs iskra-openclaw boundary, North Star. - `01-specify.md` — 11 FRs, 6 NFRs, 5 open questions, **8 falsifiable success criteria**, explicit out-of-scope. - `02-plan.md` — 4 slices (a/b/c/d), M1-M5 milestones, 6 risks + mitigations, rollback strategy per slice, composability proof. - `03-tasks.md` — atomic task list, ~2200 LOC v0 estimate, codex execution notes (one slice = one PR per ADR-0017, adversarial tests mandatory). - `04-implement-notes.md` — 13 pre-impl decisions (D1-D13: YAML over JSON, ULID over UUID, JSONL over SQLite, lazy TTL, out-of-band approval, operator-only critical tier, composability over completeness, single binary, per-session ssh-agent, wrapper-script for command capabilities, cousin-PAT for low-tier, runtime state non-sacred, ...), cross-reference matrix, known unknowns, codex pickup hints. - `README.md` — index + per-cousin reader guide. ### Code skeleton (py_compile clean) `control-plane/platformctl/agent_access/`: - `_types.py` — Capability / Session / AuditEntry / SessionStatus / AuditLevel / BackingType / CapabilityBacking dataclasses - `catalog.py` — `load_catalog`, `validate_catalog`, `lookup_capability` stubs with `TODO(codex Slice a)` markers pointing to specific 03-tasks.md entries ### Schema + seed catalog - `schema/agent-capability.schema.json` — JSON Schema draft 2020-12, validates schema_version + capabilities array with type-discriminated backing (ssh-agent / wrapped-command / infisical-token / forgejo-pat). - `policies/agent-capabilities.yaml` — 5 seed capabilities covering low/medium/high/critical audit tiers: - `forgejo-pat-read` (low, all reviewer cousins) - `ssh-rs2000-platform-host-agent` (high, codex-only, deploy operations) - `ssh-vps1000-iskra-readonly` (high, codex+claude, inspection only) - `infisical-secrets-read-scoped` (medium, scoped path) - `break-glass-full-access` (critical, operator-only via grant) ### Contract tests `control-plane/platformctl/tests/agent_access/test_catalog.py` — 10 test scaffolds (xfail strict=False) covering load/validate/lookup interface contract. Codex flips xfail off when implementing Slice (a). --- ## Cross-references - **ADR-0003** Agent Access Plane (parent) — to be reconciled at M5 - **ADR-0005** coordination lanes, **ADR-0010** cousin role taxonomy, **ADR-0017** no stacked PRs (one slice = one PR), **ADR-0018** fix root causes not workarounds (no "accept workaround" outcomes) - Composes with: **#56, #73, #74, #132, #198, #199, #200** - Supersedes: `pdurlej/iskra-openclaw#43` canonical credential architecture --- ## Tier This PR: **Trivial** (draft scaffold, no runtime/schema/sacred-path mutation, py_compile clean stubs). Future slice PRs: **Lite** (Slice a, catalog schema), **Full** (Slices b/c/d — security-sensitive, sacred-path-adjacent). --- ## Operator action Review scope. If accept → mark this PR `draft`-cleared OR close + signal codex to start Slice (a) in a fresh PR (per ADR-0017 no-stacked-prs, the Spec Kit + impl are separate PRs). Refs #76

claude added 1 commit

2026-05-17 10:17:51 +02:00

WIP: feat(agent-access): v0 4-slice skeleton — session-scoped credential delivery [#76 ]

patchwarden-pr-sanity / collect-diff (pull_request) Successful in 4s

Details

python-ci / Python 3.11 (pull_request) Failing after 21s

Details

python-ci / Python 3.12 (pull_request) Failing after 23s

Details

python-ci / Python 3.13 (pull_request) Failing after 22s

Details

patchwarden-pr-sanity / sanity (pull_request) Successful in 21s

Details

base-is-main / guard (pull_request) Successful in 2s

Details

canary-required / collect-diff (pull_request) Successful in 5s

Details

platformctl plan / auto-apply scope (pull_request) Successful in 22s

Details

pyfallow / Pyfallow gate (control-plane) (pull_request) Successful in 17s

Details

canary-required / canary (pull_request) Successful in 14s

Details

bed57fd088

Pre-impl skeleton + Spec Kit for issue #76 Agent Access Plane.

Status: draft — for review, NOT for merge. Pattern: hybrid GitHub Spec Kit +
code skeleton, same as pdurlej/iskra-openclaw PR #276/#278/#279/#281.

Author: claude. Implementer: codex (anticipated). Reviewers: claude + glm +
optionally DeepSeek per ADR-0007 Full tier.

What's in here (~1300 LOC):

Spec Kit (6 markdowns, ~1000 lines):
- docs/specs/agent-access-plane-v0/00-constitution.md — 8 principles, 10
  forbidden anti-patterns, platform vs iskra-openclaw boundary, North Star.
- 01-specify.md — 11 FRs, 6 NFRs, 5 open questions, 8 falsifiable success
  criteria, explicit out-of-scope list.
- 02-plan.md — 4 slices (a/b/c/d), M1-M5 milestones, 6 risks + mitigations,
  rollback strategy per slice, composability proof for future work.
- 03-tasks.md — atomic task list per slice, LOC estimates (~2200 LOC v0 total),
  codex execution notes per ADR-0007/0017/0018.
- 04-implement-notes.md — 13 pre-impl decisions (D1-D13), cross-reference
  matrix, known unknowns, codex pickup hints.
- README.md — index + per-cousin reader guide.

Code skeleton (py_compile clean):
- control-plane/platformctl/agent_access/__init__.py — package marker
- _types.py — Capability, Session, AuditEntry, SessionStatus, AuditLevel,
  BackingType, CapabilityBacking dataclasses
- catalog.py — load_catalog, validate_catalog, lookup_capability stubs with
  TODO(codex) markers pointing to specific 03-tasks.md entries

Schema + seed catalog:
- schema/agent-capability.schema.json — JSON Schema draft 2020-12, validates
  schema_version + capabilities array with type-discriminated backing
- policies/agent-capabilities.yaml — 5 seed capabilities: forgejo-pat-read
  (low audit), ssh-rs2000-platform-host-agent (high, codex-only), ssh-vps1000-
  iskra-readonly (high, codex+claude), infisical-secrets-read-scoped (medium),
  break-glass-full-access (critical, operator-only)

Contract tests (xfail until Slice (a) ships):
- control-plane/platformctl/tests/agent_access/test_catalog.py — 10 test
  scaffolds covering load/validate/lookup interface contract

Cross-references:
- ADR-0003 (parent Agent Access Plane ADR) — to be reconciled at M5
- ADR-0005 coordination lanes
- ADR-0007 risk-proportional canary — applied to slice tier classification
- ADR-0010 cousin role taxonomy
- ADR-0017 no stacked PRs — one slice = one PR
- ADR-0018 fix root causes not workarounds — no "accept workaround" outcomes
- Composes with: #56, #73, #74, #132, #198, #199, #200
- Supersedes: pdurlej/iskra-openclaw#43 canonical credential architecture

Tier: this PR is Trivial (draft scaffold). Future slice PRs are Lite (Slice a)
or Full (Slices b/c/d) per ADR-0007.

Refs #76

glm referenced this pull request from a commit

2026-05-17 10:24:33 +02:00

WIP: feat(wake-bus): v0 4-slice skeleton — comment-driven cousin orchestration [#134]

claude referenced this pull request

2026-05-17 10:25:03 +02:00

WIP: feat(wake-bus): v0 4-slice skeleton — comment-driven cousin orchestration [#134] #324

glm referenced this pull request from a commit

2026-05-17 10:29:38 +02:00

WIP: feat(openclaw-scheduler): v0 3-slice skeleton — scheduler observability + upgrade gate [#135]

claude referenced this pull request

2026-05-17 10:30:01 +02:00

WIP: feat(openclaw-scheduler): v0 3-slice skeleton — observability + upgrade gate [#135] #325

glm referenced this pull request from a commit

2026-05-17 16:35:35 +02:00

docs(prompts): codex execution prompts for 3 prebuild PRs (#323/#324/#325)

claude referenced this pull request

2026-05-17 16:35:55 +02:00

docs(prompts): codex execution prompts for 3 prebuild PRs (#323/#324/#325) #326

claude referenced this pull request

2026-05-17 17:25:50 +02:00

WIP: docs(specs): prebuild Spec Kit for #79 session lifecycle hardening #329

claude referenced this pull request

2026-05-17 21:04:53 +02:00

docs(specs): Kan MCP platform-managed v0 prebuild (#131) #345

claude referenced this pull request from a commit

2026-05-18 08:16:41 +02:00

docs(specs): #79 Agent Access session lifecycle hardening v0 prebuild

claude referenced this pull request

2026-05-18 08:17:08 +02:00

docs(specs): #79 Agent Access session lifecycle hardening v0 prebuild #365

claude referenced this pull request from a commit

2026-05-18 08:17:28 +02:00

docs(specs): prebuild for #56 Forgejo MCP identity split per agent v0

claude referenced this pull request

2026-05-18 08:17:56 +02:00

docs(specs): prebuild for #56 Forgejo MCP identity split per agent v0 #367

claude referenced this pull request from a commit

2026-05-18 08:19:14 +02:00

docs(specs): Matrix Heretic Room v0 prebuild (#112)

claude referenced this pull request

2026-05-18 08:19:34 +02:00

docs(specs): #112 Matrix Heretic room v0 prebuild #368

codex commented

2026-05-24 07:58:54 +02:00

Collaborator

Wave 0 Fork B triage — close/rewrite

Role: executor
Actor: codex
Decision: close this PR; keep #76 as the live work item.

Reasoning:

This PR is explicitly marked draft / NOT for merge.
It is currently non-mergeable and python-ci is red.
It mixes a large Spec Kit with first implementation slices, policy, schema, and tests. After ADR-0021/ADR-0022, this should be split into smaller Milestone 06 work: docs/spec first, then schema/policy, then platformctl implementation.
The valuable material remains preserved in this PR for reference, but the branch should not be revived by force-push from another agent.

Recommended rewrite path:

New docs/spec PR for #76 aligned with contexts/agent-coordination and ADR-0021.
Separate schema/policy PR for the capability catalog.
Separate implementation PR only after the spec and schema are accepted.

Spec sources read: docs/forgejo-agent-operations.md, state/roadmap/current-platform-roadmap.md, decisions/0021-ddd-bounded-contexts-for-platform-monorepo.md, decisions/0022-module-source-and-release-boundaries.md.

## Wave 0 Fork B triage — close/rewrite **Role:** executor **Actor:** codex **Decision:** close this PR; keep #76 as the live work item. Reasoning: - This PR is explicitly marked `draft` / `NOT for merge`. - It is currently non-mergeable and `python-ci` is red. - It mixes a large Spec Kit with first implementation slices, policy, schema, and tests. After ADR-0021/ADR-0022, this should be split into smaller Milestone 06 work: docs/spec first, then schema/policy, then platformctl implementation. - The valuable material remains preserved in this PR for reference, but the branch should not be revived by force-push from another agent. Recommended rewrite path: 1. New docs/spec PR for #76 aligned with `contexts/agent-coordination` and ADR-0021. 2. Separate schema/policy PR for the capability catalog. 3. Separate implementation PR only after the spec and schema are accepted. Spec sources read: `docs/forgejo-agent-operations.md`, `state/roadmap/current-platform-roadmap.md`, `decisions/0021-ddd-bounded-contexts-for-platform-monorepo.md`, `decisions/0022-module-source-and-release-boundaries.md`.

codex closed this pull request

2026-05-24 07:58:54 +02:00

codex referenced this pull request from a commit

2026-05-28 10:33:37 +02:00

WIP: feat(wake-bus): v0 4-slice skeleton — comment-driven cousin orchestration [#134]

codex referenced this pull request from a commit

2026-05-28 13:07:22 +02:00

WIP: feat(openclaw-scheduler): v0 3-slice skeleton — scheduler observability + upgrade gate [#135]