parity gate: one ledger, one test, two assertions that can fail today

.conductor/render-stamp.json

@@ -1,4 +1,4 @@
 {
-  "renderedAt": "2026-07-30T23:25:25.028Z",
-  "stateMtimeMs": 1785453924990.2876
+  "renderedAt": "2026-07-31T17:27:24.546Z",
+  "stateMtimeMs": 1785518844528.5571
 }

.conductor/state.json

@@ -1,6 +1,6 @@
 {
   "version": 1,
-  "active": null,
+  "active": "platform-parity-mechanism",
   "epics": [
     {
       "id": "pm-fix-link-flag-validation-and-update",
@@ -1149,12 +1149,13 @@
       "id": "platform-parity-mechanism",
       "title": "Build the mechanism that keeps every supported platform at parity with the Claude Code base, BEFORE any second platform exists. Three layers, mirroring how this repo already enforces documentation currency: (1) MECHANICAL -- a machine-parseable parity ledger (one markdown file with a table, OpenSpec supported-tools.md style, parsed directly by a CI test so exemptions are real config not prose) plus a conductor.test.mjs gate that enumerates Claude Code artifacts (commands/*.md, hook events in hooks/hooks.json, skills/, agents/) and FAILS CI when any lacks either a per-platform counterpart or an explicit exemption row with a reason; generic over whatever platforms the ledger lists, so it passes trivially at zero platforms and starts enforcing the moment codex is added. (2) PROCEDURAL -- a project-local skill that walks each platform directory in order, reads what changed in the Claude Code base, determines the per-platform delta, applies it, and updates the ledger; same shape as release-checklist/mintlify-doc-sync. (3) Semantic parity is explicitly NOT this epic's job -- a structural test can prove a counterpart file exists but never that it behaves the same; that gap is closed by the edd-harness sibling epic. Deliberately sequenced before codex-platform-support so the gate never has a window where drift could start unobserved.",
       "priority": "P1",
-      "status": "queued",
+      "status": "active",
       "role": "epic",
       "lane": "openspec",
       "links": [],
       "reconcileNeeded": false,
-      "parent": "multi-platform-agent-support"
+      "parent": "multi-platform-agent-support",
+      "startedAt": "2026-07-31T03:07:34.379Z"
     },
     {
       "id": "codex-platform-support",
@@ -1608,6 +1609,88 @@
           "done": false
         }
       ]
+    },
+    {
+      "id": "edd-corpus-expansion-nondeterministic-surfaces",
+      "title": "EDD corpus covers ONE scenario; expand to the judgment-dependent surfaces where drift actually hides",
+      "priority": "P2",
+      "status": "queued",
+      "role": "epic",
+      "lane": "claude-code",
+      "links": [],
+      "reconcileNeeded": false,
+      "stories": [
+        {
+          "title": "the harness existing is NOT the same as EDD being done -- corpus has 1 scenario (lane routing) against ~21 artifacts",
+          "done": false
+        },
+        {
+          "title": "inclusion criterion from the EDD design: an artifact earns a scenario when correct handling depends on agent JUDGMENT, not mechanical execution. Candidates: detour minimal-vs-substantial classification, the autonomy decision rule, the reconcile gate, hierarchy orchestration",
+          "done": false
+        },
+        {
+          "title": "each scenario must be mutation-tested -- this session found 4 of 5 guard tests passing on an unrelated rule, so a scenario that cannot fail is worse than none",
+          "done": false
+        }
+      ]
+    },
+    {
+      "id": "parity-reassessment-per-platform",
+      "title": "RECURRING: on every new platform, re-run the corpus, reassess ledger exemptions, and compare drift",
+      "priority": "P2",
+      "status": "planned",
+      "role": "epic",
+      "lane": "claude-code",
+      "links": [
+        {
+          "type": "blocks",
+          "epic": "hermes-platform-support",
+          "reason": "Hermes is the first platform that must go through this reassessment, under the ledger gate"
+        }
+      ],
+      "reconcileNeeded": false,
+      "stories": [
+        {
+          "title": "not a one-time task -- re-open this on EACH platform added. Add the platform to parity-ledger.json platforms[], fill its column, run the corpus against the blessed baseline, and record what differed",
+          "done": false
+        },
+        {
+          "title": "the ledger catches FORGOTTEN capabilities; EDD catches capabilities that exist but BEHAVE differently. Both must run per platform -- neither substitutes for the other",
+          "done": false
+        }
+      ]
+    },
+    {
+      "id": "parity-propagation-skill",
+      "title": "Procedural half of the parity model: the propagation skill, written against a real port",
+      "priority": "P3",
+      "status": "planned",
+      "role": "epic",
+      "lane": "claude-code",
+      "links": [],
+      "reconcileNeeded": false,
+      "stories": [
+        {
+          "title": "DEFERRED from platform-parity-mechanism on purpose: writing the procedure before any port has happened would infer steps from design docs rather than from what a port actually required",
+          "done": false
+        }
+      ]
+    },
+    {
+      "id": "parity-ledger-exemptions",
+      "title": "Ledger exemptions (capability genuinely unsupportable on a platform), designed against a real case",
+      "priority": "P3",
+      "status": "planned",
+      "role": "epic",
+      "lane": "claude-code",
+      "links": [],
+      "reconcileNeeded": false,
+      "stories": [
+        {
+          "title": "DROPPED from the initial ledger deliberately: with zero exemptions the 'every exemption has a reason' assertion could never fail -- the vacuous-coverage pattern that bit this session four times. Design it when Hermes hits a real cannot-support case.",
+          "done": false
+        }
+      ]
     }
   ],
   "detourStack": [],

CLAUDE.md

@@ -43,6 +43,15 @@
   (`.claude/skills/release-checklist/SKILL.md`) — repo-maintenance tooling, not part of what
   the plugin ships to users. Follow it every time `plugin.json`'s version bumps; don't
   re-derive the checklist from memory.
+- **Parity ledger.** Every file under `commands/`, `agents/`, `skills/`, `hooks/`, and
+  `.claude-plugin/` must be claimed by exactly one capability in `docs/parity-ledger.json`, and
+  every path it claims must exist. Git-ignored paths (e.g. a macOS `.DS_Store`) are skipped by
+  the walk, so local cruft can never trip this gate. `scripts/test/parity.test.mjs` enforces
+  both and fails CI otherwise. Adding a command/agent/skill file means adding it to a capability
+  in the same commit — either an existing one or a new one with its `claude-code` mechanism
+  described. Unported platforms are **absent** from `platforms[]`, never present with null
+  values; a port adds itself and fills its column as it goes. See
+  `docs/superpowers/specs/2026-07-31-platform-parity-mechanism-design.md`.
 - Engine subcommands are dispatched at the bottom of `conductor.mjs`; every new subcommand needs
   a matching command doc under `commands/` and coverage in `scripts/test/*.test.mjs`.
 - **State-transition flags are not pure functions of current state.** `reconcileNeeded` in

PROJECT.md

@@ -3,11 +3,11 @@
 > GENERATED by the `pm` plugin — do not hand-edit. Source of truth is
 > `.conductor/state.json` (ordering, detours, links) + OpenSpec `tasks.md` (stories).
 > Regenerate: `/pm:status` (or `node scripts/conductor.mjs render`).
-> Last rendered: 2026-07-30T23:25:25.015Z
+> Last rendered: 2026-07-31T17:27:24.539Z
 
 ## Now
 
-_No active epic set._
+**`platform-parity-mechanism`** — Build the mechanism that keeps every supported platform at parity with the Claude Code base, BEFORE any second platform exists. Three layers, mirroring how this repo already enforces documentation currency: (1) MECHANICAL -- a machine-parseable parity ledger (one markdown file with a table, OpenSpec supported-tools.md style, parsed directly by a CI test so exemptions are real config not prose) plus a conductor.test.mjs gate that enumerates Claude Code artifacts (commands/*.md, hook events in hooks/hooks.json, skills/, agents/) and FAILS CI when any lacks either a per-platform counterpart or an explicit exemption row with a reason; generic over whatever platforms the ledger lists, so it passes trivially at zero platforms and starts enforcing the moment codex is added. (2) PROCEDURAL -- a project-local skill that walks each platform directory in order, reads what changed in the Claude Code base, determines the per-platform delta, applies it, and updates the ledger; same shape as release-checklist/mintlify-doc-sync. (3) Semantic parity is explicitly NOT this epic's job -- a structural test can prove a counterpart file exists but never that it behaves the same; that gap is closed by the edd-harness sibling epic. Deliberately sequenced before codex-platform-support so the gate never has a window where drift could start unobserved. (epic, P1) — —
 
 ## Detour stack
 
@@ -48,11 +48,13 @@ _Empty — no work is paused._
 | P2 | `df-verify-worktrees-merged-not-just-archived` | claude-code | epic | archived 🤖 | — | - |
 | P2 | `docs-site-cleanup` | claude-code | epic | archived | 3/9 stories | - |
 | P2 | `edd-adapter-discards-agent-stdout` | claude-code | epic | queued | — | - |
+| P2 | `edd-corpus-expansion-nondeterministic-surfaces` | claude-code | epic | queued | 0/3 stories | - |
 | P2 | `edd-measures-installed-plugin-not-worktree` | claude-code | epic | queued | — | blocks→hermes-platform-support |
 | P2 | `gh-64-sync-duplicate-shipped-plan` | claude-code | epic | queued | 0/3 stories | - |
 | P2 | `gh-69-sync-no-done-signal-for-plans` | claude-code | epic | queued | — | relates-to→gh-64-sync-duplicate-shipped-plan |
 | P2 | `gh-70-link-validation-gap` | claude-code | epic | queued | — | - |
 | P2 | `github-issue-tracker-sync` | claude-code | epic | archived 🤖 | — | relates-to→ai-feedback-loop-github-issues |
+| P2 | `parity-reassessment-per-platform` | claude-code | epic | planned | 0/2 stories | blocks→hermes-platform-support |
 | P2 | `platform-switch-orphans-old-rules-block` | claude-code | epic | queued | 0/2 stories | blocks→hermes-platform-support |
 | P2 | `pm-plugin-improvements-2026-07-14` | claude-code | epic | archived | 19/19 children archived | - |
 | P0 | └─ `df-plan-hierarchy-includes-archived-children` | claude-code | epic | archived | — | - |
@@ -80,11 +82,13 @@ _Empty — no work is paused._
 | P3 | `gh-66-update-epic-missing-flags` | claude-code | epic | queued | — | blocks→gh-69-sync-no-done-signal-for-plans |
 | P3 | `init-detects-foreign-agent-instruction-files` | claude-code | epic | planned | 0/1 stories | - |
 | P3 | `multi-platform-agent-support` | claude-code | epic | queued | 2/5 children archived | - |
-| P1 | └─ `platform-parity-mechanism` | openspec | epic | queued ⚠ no change on disk | — | - |
+| P1 | └─ `platform-parity-mechanism` | openspec | epic | active ⚠ no change on disk | — | - |
 | P1 | └─ `rules-block-hardcodes-claude-slash-commands` | claude-code | epic | archived | — | - |
 | P1 | └─ `edd-harness-agent-behavior-testing` | decision | epic | archived | — | - |
 | P2 | └─ `hermes-platform-support` | openspec | epic | queued ⚠ no change on disk | — | depends-on→platform-parity-mechanism; depends-on→edd-harness-agent-behavior-testing; depends-on→rules-block-hardcodes-claude-slash-commands |
 | P3 | └─ `codex-platform-support` | openspec | epic | queued ⚠ no change on disk | — | depends-on→platform-parity-mechanism; depends-on→edd-harness-agent-behavior-testing; depends-on→rules-block-hardcodes-claude-slash-commands; depends-on→hermes-platform-support |
+| P3 | `parity-ledger-exemptions` | claude-code | epic | planned | 0/1 stories | - |
+| P3 | `parity-propagation-skill` | claude-code | epic | planned | 0/1 stories | - |
 | P3 | `rules-block-cli-fallback` | claude-code | epic | planned | 0/1 stories | - |
 | P3 | `portfolio-architecture-consistency-scan` | decision | epic | archived | — | depends-on→epic-hierarchy-orchestration |
 | P? | `2026-07-14-epic-hierarchy-orchestration` | superpowers | epic | archived | 0/17 tasks | - |
@@ -108,20 +112,23 @@ _Empty — no work is paused._
 ## Briefing (what a fresh session sees)
 
 ```
+⚠ pm 0.24.0 → 0.25.0 available — run `/reload-plugins` (if you just updated the plugin), then `/pm:upgrade`.
+   - **`rules-target` — a read-only query printing the absolute path of the file the resolved
+
 CONDUCTOR STATE — where we are and what's next
 
-NOW: (no active epic set)
+NOW: `platform-parity-mechanism` (openspec, epic, P1) — —
 
 NEXT UP (by priority, then lane):
   • `autodetour-parser-misses-am-and-f` (P2, claude-code, queued) — 0/3 stories
   • `edd-adapter-discards-agent-stdout` (P2, claude-code, queued) — —
+  • `edd-corpus-expansion-nondeterministic-surfaces` (P2, claude-code, queued) — 0/3 stories
   • `edd-measures-installed-plugin-not-worktree` (P2, claude-code, queued) — —
   • `gh-64-sync-duplicate-shipped-plan` (P2, claude-code, queued) — 0/3 stories
-  • `gh-69-sync-no-done-signal-for-plans` (P2, claude-code, queued) — —
-  (+4 more — see PROJECT.md)
-  lanes: openspec 2 · superpowers 5 · claude-code 55 · decision 7
+  (+5 more — see PROJECT.md)
+  lanes: openspec 2 · superpowers 5 · claude-code 56 · decision 7
 
-planned: 2 — see PROJECT.md
+planned: 5 — see PROJECT.md
 
 EPIC LINKS:
   • `edd-observe-hardcodes-claude-md` depends-on `rules-block-hardcodes-claude-slash-commands` — observe.py correctly observes what the engine WRITES; the engine hardcodes CLAUDE.md (constants.mjs:14) and writes only there, so parameterizing observe first would abstract over a filename nothing produces
@@ -133,6 +140,7 @@ EPIC LINKS:
   • `edd-measures-installed-plugin-not-worktree` blocks `hermes-platform-support` — editing an evaluated artifact must be measurable before the port, and the baseline must record which plugin version it describes
   • `gh-69-sync-no-done-signal-for-plans` relates-to `gh-64-sync-duplicate-shipped-plan` — same root cause -- dedup keys on the plan's filename-derived id instead of a plan<->epic association, so any epic whose plan is named differently is re-registered forever
   • `github-issue-tracker-sync` relates-to `ai-feedback-loop-github-issues` — issues created by the feedback loop are exactly what this syncs back in
+  • `parity-reassessment-per-platform` blocks `hermes-platform-support` — Hermes is the first platform that must go through this reassessment, under the ledger gate
   • `platform-switch-orphans-old-rules-block` blocks `hermes-platform-support` — a switched repo would leave stale conductor instructions in the file the previous platform read
   • `codex-platform-support` depends-on `platform-parity-mechanism` — the structural parity gate and ledger must exist before any second-platform port
   • `codex-platform-support` depends-on `edd-harness-agent-behavior-testing` — semantic parity needs a blessed Claude Code baseline to compare against
@@ -142,7 +150,7 @@ EPIC LINKS:
   • `portfolio-architecture-consistency-scan` depends-on `epic-hierarchy-orchestration` — C presumes B's hierarchy execution exists to propagate cross-cutting decisions into
 
 TRACKER SYNC (github-issues):
-  ⚠ not yet in github-issues — create issues + record keys (update-epic): `autodetour-parser-misses-am-and-f`, `edd-adapter-discards-agent-stdout`, `edd-measures-installed-plugin-not-worktree`, `platform-switch-orphans-old-rules-block`, `multi-platform-agent-support`
+  ⚠ not yet in github-issues — create issues + record keys (update-epic): `autodetour-parser-misses-am-and-f`, `edd-adapter-discards-agent-stdout`, `edd-corpus-expansion-nondeterministic-surfaces`, `edd-measures-installed-plugin-not-worktree`, `platform-switch-orphans-old-rules-block`, `multi-platform-agent-support`
 
 💡 1 tracker configured (github-issues) — consider `/pm:sync` this session to pull in any new issues.
 

docs/parity-ledger.json

@@ -0,0 +1,73 @@
+{
+  "platforms": ["claude-code"],
+  "capabilities": [
+    {
+      "id": "conductor-discipline",
+      "artifacts": ["skills/conductor/SKILL.md"],
+      "platforms": { "claude-code": "skill loaded by name (skills/<name>/SKILL.md)" }
+    },
+    {
+      "id": "epic-index",
+      "artifacts": [
+        "commands/epic.md",
+        "commands/status.md",
+        "commands/next.md",
+        "commands/sync.md"
+      ],
+      "platforms": { "claude-code": "slash commands (/pm:epic, /pm:status, /pm:next, /pm:sync)" }
+    },
+    {
+      "id": "detour-lifecycle",
+      "artifacts": [
+        "commands/detour.md",
+        "commands/resume.md",
+        "commands/gate-guard.md",
+        "agents/reconciler.md"
+      ],
+      "platforms": { "claude-code": "slash commands + reconciler subagent + a PreToolUse gate-guard hook" }
+    },
+    {
+      "id": "epic-hierarchy-orchestration",
+      "artifacts": [
+        "commands/hierarchy.md",
+        "agents/hierarchy-child-executor.md",
+        "agents/merge-conflict-resolver.md"
+      ],
+      "platforms": { "claude-code": "slash command dispatching two subagents into git worktrees" }
+    },
+    {
+      "id": "repo-configuration",
+      "artifacts": [
+        "commands/tracker.md",
+        "commands/lane-routing.md",
+        "commands/review-mode.md"
+      ],
+      "platforms": { "claude-code": "slash commands writing settings into .conductor/state.json" }
+    },
+    {
+      "id": "install-and-upgrade",
+      "artifacts": [
+        "commands/init.md",
+        "commands/upgrade.md",
+        "commands/changelog.md",
+        "commands/changesets.md"
+      ],
+      "platforms": { "claude-code": "slash commands (/pm:init scaffolds; /pm:upgrade migrates and rewrites the rules block)" }
+    },
+    {
+      "id": "feedback-channel",
+      "artifacts": ["commands/feedback.md"],
+      "platforms": { "claude-code": "slash command instructing the agent to file a GitHub issue (the engine never calls gh)" }
+    },
+    {
+      "id": "session-lifecycle-hooks",
+      "artifacts": ["hooks/hooks.json"],
+      "platforms": { "claude-code": "SessionStart / PreCompact / PostToolUse hooks in hooks.json, each passing --platform claude-code" }
+    },
+    {
+      "id": "plugin-packaging",
+      "artifacts": [".claude-plugin/plugin.json"],
+      "platforms": { "claude-code": "plugin manifest consumed by the Claude Code marketplace loader" }
+    }
+  ]
+}