fix(agents): mode cards load their mode-protocol skill at kickoff (#989)

packages/extension/agents/develop.md

@@ -14,11 +14,14 @@ the shipped skill library). You automate the *walk*, never the *gate*:
 every package edit still requires an issue and a PR, CI green is still the merge
 condition, and promotions stay human-only.
 
-**First action (kickoff or resume): invoke the `director-core` skill and follow it.**
-It is the canonical loop protocol; the spine below is its summary, never a
-replacement. Your mode's specifics — the phase graph, gates, and roles — are the
-**dev gate pack** (`modes/develop/pack.toml` in the amicode repo, schema'd and
-fixture-tested): phases decompose → implement → integrate.
+**First actions (kickoff or resume): invoke the `director-core` skill, THEN the
+`autodev` skill — both, before acting on the loop.** `director-core` is the
+canonical, engine-neutral loop protocol; `autodev` is this mode's binding of it
+(the gate pack's phases, the implementer cast, the dispatch and gate discipline) —
+the spine below is a summary of both, never a replacement for either. Your mode's
+specifics — the phase graph, gates, and roles — are the **dev gate pack**
+(`modes/develop/pack.toml` in the amicode repo, schema'd and fixture-tested):
+phases decompose → implement → integrate.
 
 ## The spine
 

packages/extension/agents/research.md

@@ -14,10 +14,13 @@ principle is fixed: **the context window is a cache; the vault is the database.*
 piece of load-bearing state lives in vault notes; the context holds only the working set;
 any compaction costs a cache refill, never state.
 
-**First action (kickoff or resume): invoke the `director-core` skill and follow it.** It
-is the canonical loop protocol; the spine below is its summary, never a replacement.
-Your mode's specifics — the phase graph, gates, and roles — are the **research gate
-pack** (`modes/research/pack.toml` in the amicode repo, schema'd and fixture-tested):
+**First actions (kickoff or resume): invoke the `director-core` skill, THEN the
+`research` skill — both, before acting on the loop.** `director-core` is the
+canonical, engine-neutral loop protocol; `research` is this mode's binding of it
+(the loop steps, the spec-gate mechanics, the roles, the probe/experiment boundary)
+— the spine below is a summary of both, never a replacement for either. Your mode's
+specifics — the phase graph, gates, and roles — are the **research gate pack**
+(`modes/research/pack.toml` in the amicode repo, schema'd and fixture-tested):
 phases hypothesize → deliberate → experiment → gate → analyze.
 
 ## The spine

packages/extension/modes/develop/card.md

@@ -14,11 +14,14 @@ the shipped skill library). You automate the *walk*, never the *gate*:
 every package edit still requires an issue and a PR, CI green is still the merge
 condition, and promotions stay human-only.
 
-**First action (kickoff or resume): invoke the `director-core` skill and follow it.**
-It is the canonical loop protocol; the spine below is its summary, never a
-replacement. Your mode's specifics — the phase graph, gates, and roles — are the
-**dev gate pack** (`modes/develop/pack.toml` in the amicode repo, schema'd and
-fixture-tested): phases decompose → implement → integrate.
+**First actions (kickoff or resume): invoke the `director-core` skill, THEN the
+`autodev` skill — both, before acting on the loop.** `director-core` is the
+canonical, engine-neutral loop protocol; `autodev` is this mode's binding of it
+(the gate pack's phases, the implementer cast, the dispatch and gate discipline) —
+the spine below is a summary of both, never a replacement for either. Your mode's
+specifics — the phase graph, gates, and roles — are the **dev gate pack**
+(`modes/develop/pack.toml` in the amicode repo, schema'd and fixture-tested):
+phases decompose → implement → integrate.
 
 ## The spine
 

packages/extension/modes/research/card.md

@@ -14,10 +14,13 @@ principle is fixed: **the context window is a cache; the vault is the database.*
 piece of load-bearing state lives in vault notes; the context holds only the working set;
 any compaction costs a cache refill, never state.
 
-**First action (kickoff or resume): invoke the `director-core` skill and follow it.** It
-is the canonical loop protocol; the spine below is its summary, never a replacement.
-Your mode's specifics — the phase graph, gates, and roles — are the **research gate
-pack** (`modes/research/pack.toml` in the amicode repo, schema'd and fixture-tested):
+**First actions (kickoff or resume): invoke the `director-core` skill, THEN the
+`research` skill — both, before acting on the loop.** `director-core` is the
+canonical, engine-neutral loop protocol; `research` is this mode's binding of it
+(the loop steps, the spec-gate mechanics, the roles, the probe/experiment boundary)
+— the spine below is a summary of both, never a replacement for either. Your mode's
+specifics — the phase graph, gates, and roles — are the **research gate pack**
+(`modes/research/pack.toml` in the amicode repo, schema'd and fixture-tested):
 phases hypothesize → deliberate → experiment → gate → analyze.
 
 ## The spine

packages/extension/test/mode_cards.test.ts

@@ -237,6 +237,41 @@ describe("mode cards — frontmatter (boot-check fields)", () => {
   }
 });
 
+// ── #989: first-actions skill pairing ────────────────────────────────────────
+// Each mode card must wire the director into BOTH the shared spine AND its
+// own mode-protocol skill at kickoff/resume, in that order. The 2026-09-10
+// research pickup loaded only director-core (the directive named no mode
+// skill, so the director treated the spine summary as sufficient and skipped
+// the mode binding — spec-gate mechanics, roles, probe/experiment boundary);
+// the cards now name both, and this floor pins the pairing.
+const MODE_SKILL: Record<CardName, string> = {
+  // the dev mode's protocol skill (its id keeps the pre-rename mode name)
+  "develop.md": "autodev",
+  // the research mode's protocol skill (ex-autoresearch; old id read-resolves)
+  "research.md": "research",
+};
+
+describe("mode cards — first-actions skill pairing (#989)", () => {
+  for (const name of CARDS) {
+    it(`${name}: first-action directive loads director-core AND its mode skill`, () => {
+      const text = cardText(name);
+      // the directive lives in the preamble, before the delimited spine
+      const preamble = text.slice(0, text.indexOf(SPINE_START));
+      const coreAt = preamble.indexOf("`director-core`");
+      const modeAt = preamble.indexOf(`\`${MODE_SKILL[name]}\``);
+      expect(
+        coreAt,
+        "the card names the shared spine skill in its first-action directive",
+      ).toBeGreaterThan(-1);
+      expect(
+        modeAt,
+        `the card names its mode-protocol skill (\`${MODE_SKILL[name]}\`) in its first-action directive`,
+      ).toBeGreaterThan(-1);
+      expect(modeAt, "the mode skill is named after the spine skill").toBeGreaterThan(coreAt);
+    });
+  }
+});
+
 describe("mode cards — blocklist (open-protocol vocabulary)", () => {
   for (const name of CARDS) {
     it(`${name}: zero blocklisted proprietary strings`, () => {