docs(specs): #131 Kan MCP platform-managed discovery v0 prebuild #366

Merged

pdurlej merged 1 commit from claude/fork-B2-kan-mcp-platform-managed into main

2026-05-23 09:19:37 +02:00

claude commented

2026-05-18 08:17:38 +02:00

Collaborator

Summary

Fork-B2 of 2026-05-18 retry batch (per dispatch template PR #356, worktree-isolated to prevent shared-checkout collision lesson from 2026-05-17).

Prebuild Spec Kit + companion codex execution prompt for migrating OpenClaw's discovery of Kan MCP from a live-only systemd drop-in into platform-managed configuration via mcporter + Infisical. Closes the operator-readable drift named in issue #131.

What's in

docs/specs/kan-mcp-platform-managed-v0/ (6 files):

README.md — overview, TL;DR scope, cross-repo boundary
00-constitution.md — 9 non-negotiable principles, non-goals, stop conditions
01-specify.md — acceptance criteria sections A-G (machine-checkable + operator-verifiable)
02-plan.md — 9 design decisions with rejected alternatives, tier classification
03-tasks.md — per-slice task breakdown (Slice a/b/c, plus d out of v0)
04-implement-notes.md — 7 gotchas, prior art, wiring, anti-patterns

prompts/codex-kan-mcp-platform-managed.md — execution runbook with Safety/production boundary section per Codex feedback + PR #326 convention.

Key principles

P1 Source of truth in platform repo, not on host
P2 Kan agent token never reaches OpenClaw (stays at bridge on RS2000)
P5 Backwards compat during rollout; drop-in stays until separate Slice d
P6 Composes with PR #349 (secrets-pipeline-foundation-v0) — same Infisical scope expansion that #131 originally blocked on (403 create on secrets)
P8 Renderer output stability — mcp.servers.kan structure identical post-migration

Slice tiers (per ADR-0007)

Slice (a) registry + module dep — Lite, self-merge OK
Slice (b) Infisical scope expansion — Full, class/security-sensitive, operator merge; depends on PR #349
Slice (c) cross-repo coordination with pdurlej/iskra-openclaw — Full, issue-based (not PR-stacking); operator merge
Slice (d) drop-in removal — OUT OF V0 SCOPE

Per ADR-0018

Explicit anti-patterns + stop conditions reject "accept workaround" outcomes. Constitution P2 (token boundary) non-negotiable. Future cousins reading the drop-in as fire-signal is exactly the failure mode this Spec Kit fixes at root.

Tier (THIS PR)

Trivial per ADR-0007 — prebuild docs, no runtime/schema/sacred-path mutation. Implementation slices tiered separately.

Estimated v0 effort

5-7h wall-clock across 3 PR cycles (Slice a → b → c).

Refs #131, PR #349 (sibling Infisical work), PR #356 (fork dispatch template)

## Summary Fork-B2 of 2026-05-18 retry batch (per dispatch template PR #356, worktree-isolated to prevent shared-checkout collision lesson from 2026-05-17). Prebuild Spec Kit + companion codex execution prompt for migrating OpenClaw's discovery of Kan MCP from a live-only systemd drop-in into platform-managed configuration via mcporter + Infisical. Closes the operator-readable drift named in issue #131. ## What's in `docs/specs/kan-mcp-platform-managed-v0/` (6 files): - README.md — overview, TL;DR scope, cross-repo boundary - 00-constitution.md — 9 non-negotiable principles, non-goals, stop conditions - 01-specify.md — acceptance criteria sections A-G (machine-checkable + operator-verifiable) - 02-plan.md — 9 design decisions with rejected alternatives, tier classification - 03-tasks.md — per-slice task breakdown (Slice a/b/c, plus d out of v0) - 04-implement-notes.md — 7 gotchas, prior art, wiring, anti-patterns `prompts/codex-kan-mcp-platform-managed.md` — execution runbook with Safety/production boundary section per Codex feedback + PR #326 convention. ## Key principles - **P1** Source of truth in platform repo, not on host - **P2** Kan agent token never reaches OpenClaw (stays at bridge on RS2000) - **P5** Backwards compat during rollout; drop-in stays until separate Slice d - **P6** Composes with PR #349 (secrets-pipeline-foundation-v0) — same Infisical scope expansion that #131 originally blocked on (`403 create on secrets`) - **P8** Renderer output stability — `mcp.servers.kan` structure identical post-migration ## Slice tiers (per ADR-0007) - **Slice (a)** registry + module dep — Lite, self-merge OK - **Slice (b)** Infisical scope expansion — Full, `class/security-sensitive`, operator merge; depends on PR #349 - **Slice (c)** cross-repo coordination with `pdurlej/iskra-openclaw` — Full, issue-based (not PR-stacking); operator merge - **Slice (d)** drop-in removal — OUT OF V0 SCOPE ## Per ADR-0018 Explicit anti-patterns + stop conditions reject "accept workaround" outcomes. Constitution P2 (token boundary) non-negotiable. Future cousins reading the drop-in as fire-signal is exactly the failure mode this Spec Kit fixes at root. ## Tier (THIS PR) **Trivial** per ADR-0007 — prebuild docs, no runtime/schema/sacred-path mutation. Implementation slices tiered separately. ## Estimated v0 effort 5-7h wall-clock across 3 PR cycles (Slice a → b → c). Refs #131, PR #349 (sibling Infisical work), PR #356 (fork dispatch template)

claude added 1 commit

2026-05-18 08:17:38 +02:00

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

base-is-main / guard (pull_request) Successful in 1s

Details

canary-required / collect-diff (pull_request) Successful in 3s

Details

patchwarden-pr-sanity / collect-diff (pull_request) Successful in 3s

Details

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

Details

patchwarden-pr-sanity / sanity (pull_request) Successful in 20s

Details

8dc054fdeb

Prebuild Spec Kit + companion codex execution prompt for migrating
OpenClaw's discovery of Kan MCP from a live-only systemd drop-in
(/home/openclaw/.config/systemd/user/openclaw-gateway.service.d/kan-mcp.conf)
into platform-managed configuration via mcporter + Infisical.

What's in the Spec Kit (docs/specs/kan-mcp-platform-managed-v0/):
- README.md — overview, TL;DR scope, cross-repo boundary, references
- 00-constitution.md — 9 non-negotiable principles + non-goals + stop conditions
- 01-specify.md — acceptance criteria (machine-checkable + operator-verifiable)
- 02-plan.md — 9 design decisions with alternatives rejected; tier classification
- 03-tasks.md — per-slice task breakdown (Slice a/b/c, plus d out of v0)
- 04-implement-notes.md — 7 gotchas, prior art refs, wiring notes, anti-patterns

Companion: prompts/codex-kan-mcp-platform-managed.md (with Safety/production
boundary section per Codex 2026-05-17 feedback + per PR #326 convention).

Key principles:
- Source of truth in pdurlej/platform repo, not on host (P1)
- Kan agent token never reaches OpenClaw — stays at bridge on RS2000 (P2)
- Tailnet-only Kan write surface preserved (P3)
- Backwards compat during rollout; drop-in stays until Slice d gate (P5)
- Composes with secrets-pipeline-foundation-v0 (PR #349) — same Infisical
  scope expansion that #131 originally blocked on (P6)
- No code in OpenClaw runtime path within this PR boundary (P7)
- Renderer output stability (mcp.servers.kan structure identical) (P8)

Slice classification per ADR-0007:
- Slice (a) registry + module dep — Lite tier, self-merge OK
- Slice (b) Infisical scope expansion — Full tier, class/security-sensitive,
  operator merge; depends on #349
- Slice (c) cross-repo coordination with pdurlej/iskra-openclaw — Full tier,
  cross-repo issue (not PR-stacking); operator merge
- Slice (d) drop-in removal — OUT OF V0 SCOPE; future Slice + observation
  window required

Per ADR-0018: explicit anti-patterns documented; stop conditions reject
"accept workaround" outcomes; Constitution P2 (token boundary) is
non-negotiable across all slices.

Per Codex 2026-05-17 feedback (PR #326): execution prompt includes Safety/
production boundary section + per-slice instructions + stop conditions +
reporting format + cousin coordination + references.

Estimated v0 effort: 5-7h wall-clock across 3 PR cycles (Slice a/b/c).

Tier: Trivial per ADR-0007 (prebuild file, no runtime/schema/sacred-path
mutation in THIS PR — implementation slices are tiered separately).

Fork-B2 of 2026-05-18 retry batch (per dispatch template PR #356).
Worktree-isolated to prevent shared-checkout collision (lesson from
2026-05-17).

Refs #131, PR #349, PR #356