@ericrisco/rsc 0.1.28 → 0.1.30

package/manifest.json

@@ -1,5 +1,5 @@
 {
-  "version": "0.1.28",
+  "version": "0.1.30",
   "counts": {
     "skills": 231
   },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ericrisco/rsc",
-  "version": "0.1.28",
+  "version": "0.1.30",
   "description": "Eric Risco's agent-skills catalog as a granular, self-recommending CLI installer.",
   "type": "module",
   "bin": {

package/scripts/install-apply.js

@@ -4,6 +4,7 @@ import { join, dirname } from 'node:path';
 import { fileURLToPath } from 'node:url';
 import { planInstall } from './install-plan.js';
 import { targetPaths, writeSkill, wireHook, unwireHook, baseDir, TARGET_IDS } from '../targets/index.js';
+import { targetHasAgents, writeDeveloperAgent, removeDeveloperAgent, developerAgentPath } from '../targets/agents.js';
 import { readState, writeState } from './lib/state.js';
 import { createBackup } from './lib/backups.js';
 
@@ -62,6 +63,7 @@ function managedPathsForInstall({ skillIds, target, home, cwd }) {
   const paths = targetPaths(target, home, cwd);
   const plan = planInstall({ skillIds, target, home, cwd });
   const out = [paths.stateFile, versionFile(cwd), baseVersionsFile(cwd)];
+  if (targetHasAgents(target)) out.push(developerAgentPath(target, cwd), join(cwd, '.rsc', 'developer.json'));
   for (const step of plan) {
     if (step.kind === 'skill') {
       out.push(step.to, baseDir(step.id, cwd));
@@ -94,6 +96,11 @@ export async function applyInstall({ skillIds, target, home, cwd = process.cwd()
     }
   }
   writeBaseVersions(cwd, baseVersions);
+  // Install the `developer` subagent for targets that support file-based agents (Claude
+  // Code, Cursor, OpenCode, Gemini, Copilot, Junie, Kiro, Codex). It runs at the balanced
+  // tier by default (never light/Haiku); the tier lives in .rsc/developer.json, set by
+  // `init` at onboarding and honored on every (re)install/sync.
+  if (targetHasAgents(target)) state.agents = writeDeveloperAgent(target, cwd).length ? ['developer'] : [];
   state.version = CLI_VERSION;
   writeState(paths.stateFile, state);
   mkdirSync(dirname(versionFile(cwd)), { recursive: true });
@@ -174,6 +181,9 @@ export async function purge({ home, cwd = process.cwd(), withDocs = false, dryRu
     }
     // unwireHook mutates files, so only run it for real (dry runs skip it).
     if (!dryRun) removed.push(...unwireHook(target, paths));
+    // Remove the installed developer subagent (agent-capable targets only).
+    const agentFile = developerAgentPath(target, cwd);
+    if (agentFile) drop(agentFile);
   }
   drop(join(cwd, '.rsc'), true);
   if (withDocs) drop(join(cwd, '02-DOCS'), true);

package/skills/agent-eval/SKILL.md

@@ -199,7 +199,7 @@ Pick by where the eval runs and what it must do. Versions as of 2026-06 — re-v
 
 ## Project grounding
 
-If the workspace has a `02-DOCS/` harness, record the eval policy in `02-DOCS/wiki/stack/evals.md`: dataset location, scorer mix, gate baseline file, judge model, and the failure modes covered. Link it from root `CLAUDE.md` under `## Knowledge map`. This is **recorded, not gated** — skip silently if there is no harness.
+If the workspace has a `02-DOCS/` harness, record the eval policy in `02-DOCS/wiki/stack/evals.md`: dataset location, scorer mix, gate baseline file, judge model, and the failure modes covered. Index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer). This is **recorded, not gated** — skip silently if there is no harness.
 
 ## verify.sh
 

package/skills/analyze/SKILL.md

@@ -53,7 +53,7 @@ If no profile exists, default to L2 and note that the harness has not gauged the
 
 ## Inputs — locate the four artifacts
 
-Read all four before analyzing. The rsc-sdd artifacts live under `02-DOCS/wiki/sdd/` (the harness Karpathy-wiki convention), indexed from the root `CLAUDE.md` `## Knowledge map`:
+Read all four before analyzing. The rsc-sdd artifacts live under `02-DOCS/wiki/sdd/` (the harness Karpathy-wiki convention), indexed from `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it):
 
 | Artifact | Canonical location | Role in the check |
 | --- | --- | --- |
@@ -111,7 +111,7 @@ Produce a single consistency report:
 3. **Findings table** — `# | Severity | Type | Artifact A (loc) | Artifact B (loc) | Conflict | Resolve in (phase)`.
 4. **Recommended routing** — group fixes by the phase that owns them (`clarify` for spec ambiguity, `plan` for missing architecture, `tasks` for a missing done-check, `constitution` if a principle itself is wrong).
 
-Write the report to `02-DOCS/wiki/sdd/analysis/<slug>.md` (create the dir if absent) and add/refresh its row in the root `CLAUDE.md` `## Knowledge map` under the `sdd/` topic, so the next phase and the harness can find it. The report is the artifact analyze owns — it is the *only* thing analyze writes. Per-run point-in-time; overwrite on re-run, the wiki keeps history.
+Write the report to `02-DOCS/wiki/sdd/analysis/<slug>.md` (create the dir if absent) and index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer) under the `sdd/` topic, so the next phase and the harness can find it. The report is the artifact analyze owns — it is the *only* thing analyze writes. Per-run point-in-time; overwrite on re-run, the wiki keeps history.
 
 Adapt the rendered verbosity to the dial (L0 = table only; L3 = full walk-through). Do not log a decision to `decisions.md` — analyze decides nothing; the phase that resolves the finding logs its own decision.
 

package/skills/author-skill/SKILL.md

@@ -124,7 +124,7 @@ Run in order. Each step gates the next.
 4. **Write the body** in the rsc voice (see below). Tag every code/example fence with a language. Add a copy-able checklist or decision table *only where the flow actually branches* — not as decoration. Add a short anti-patterns table.
 5. **Extract references** for anything long or branch-specific.
 6. **Write the evals** — `cases.yaml` then `README.md`. → `references/eval-authoring.md`.
-7. **Wire it into the rsc plumbing** (`tags`, `recommends`, `npm run manifest`, Knowledge map). → `references/rsc-conventions.md`.
+7. **Wire it into the rsc plumbing** (`tags`, `recommends`, `npm run manifest`, and indexing any artifact in `02-DOCS/wiki/index.md` — the Knowledge map; root `CLAUDE.md` keeps only a short pointer). → `references/rsc-conventions.md`.
 8. **Self-audit against the rubric** (below). Fix every miss or justify it.
 
 ## The rsc voice
@@ -149,7 +149,7 @@ A skill ships only when every box is checked or a miss is consciously justified.
 - [ ] **Every fence language-tagged**; no placeholder/TODO prose; examples concrete.
 - [ ] **Checklist/decision table only where a flow branches**; an **anti-patterns table** present.
 - [ ] **Accompaniment dial honored** — reads the profile, adapts verbosity.
-- [ ] **Artifacts under `02-DOCS/wiki/`** and indexed in the root CLAUDE.md Knowledge map, if the skill produces any.
+- [ ] **Artifacts under `02-DOCS/wiki/`** and indexed in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer), if the skill produces any.
 - [ ] **Concrete tooling delegated** to the stack skills rather than reinvented.
 - [ ] **evals present** — `cases.yaml` (≥5 `should_trigger` incl. non-obvious, ≥4 `should_not_trigger` each with a real-sibling `route_to`, ≥1 `capability` with a `must_include` rubric) + an honest `README.md`. `scripts/eval-lint.sh` passes — but it only checks presence and the counts (≥5/≥4/≥1) and that those keys are lists; the `route_to`-points-at-a-real-sibling, non-obvious phrasings, and `must_include` quality are yours to verify here, not the linter's.
 - [ ] **verify.sh** present iff the skill has a checkable artifact; process skills rely on evals.
@@ -195,7 +195,7 @@ only when **both** are green:
 
 ## Project grounding (02-DOCS + CLAUDE.md)
 
-When authoring produces a durable design note (a skill's scope decision, a description rationale worth keeping), persist it under `02-DOCS/wiki/sdd/` and add a row to the root CLAUDE.md `## Knowledge map`, per the `../harness/SKILL.md` convention — never a stray file at the repo root. The skill's own `evals/` is the executable record of intent; the wiki note is the human-readable why.
+When authoring produces a durable design note (a skill's scope decision, a description rationale worth keeping), persist it under `02-DOCS/wiki/sdd/` and index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer), per the `../harness/SKILL.md` convention — never a stray file at the repo root. The skill's own `evals/` is the executable record of intent; the wiki note is the human-readable why.
 
 ## See Also
 

package/skills/building-agents/SKILL.md

@@ -449,14 +449,12 @@ if __name__ == "__main__":
 
 When this skill runs in a project with a `02-DOCS/` layer (the
 [`harness`](../harness/SKILL.md) Karpathy wiki), record this
-project's agent decisions there and index them from the root `CLAUDE.md`, so the next
+project's agent decisions there and index them in `02-DOCS/wiki/index.md`, so the next
 agent inherits the conventions instead of re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/agents.md`, linked from a `## Knowledge map` section in the root
-   `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/agents.md`, indexed in `02-DOCS/wiki/index.md` (the Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — the provider(s) and model routing, where the provider adapter lives, tool/RAG conventions, the eval gates, and the observability backend —
-   then add/refresh the `CLAUDE.md` link (create the `## Knowledge map` section, and
-   `CLAUDE.md` itself, if absent).
+   then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the
    article (bump its `Updated` date) in the same change.
 

package/skills/constitution/SKILL.md

@@ -13,7 +13,7 @@ origin: risco
 
 A constitution is small, durable, and enforceable. It is **not** a wiki of everything you know about the project (that is what `02-DOCS/wiki/` already is, run by the `harness`). It is the short list of principles that, if violated, mean the work is wrong regardless of whether it runs. If a rule here cannot be checked or pointed at later, it does not belong here — move it to the stack wiki and link it.
 
-This skill produces `02-DOCS/wiki/sdd/constitution.md` and one Knowledge-map row in the root `CLAUDE.md`. It reconciles with — never duplicates — the stack conventions the harness keeps under `02-DOCS/wiki/stack/*`.
+This skill produces `02-DOCS/wiki/sdd/constitution.md` and one Knowledge-map row. The constitution is one of the few **read-first** pointer entries kept directly in the root `CLAUDE.md` `## Knowledge map` (everything else lives in `02-DOCS/wiki/index.md`, the full Knowledge map that root `CLAUDE.md` points to). It reconciles with — never duplicates — the stack conventions the harness keeps under `02-DOCS/wiki/stack/*`.
 
 ## Model tier — `heavy` (opt-in routing)
 
@@ -52,7 +52,7 @@ Do NOT use when (route instead):
 
 The harness may already hold real conventions under `02-DOCS/wiki/stack/*` (e.g. `nextjs.md`, `fastapi.md`, `postgresdb.md`). The constitution does not copy them — it **ratifies the principle and links the detail**. Run this reconciliation pass first:
 
-1. **Read the Knowledge map** in the root `CLAUDE.md`. List every `02-DOCS/wiki/stack/*` article that exists.
+1. **Read the Knowledge map** — the full index lives in `02-DOCS/wiki/index.md` (root `CLAUDE.md` keeps only a short pointer to it). List every `02-DOCS/wiki/stack/*` article that exists.
 2. **Read each stack article.** Pull out anything already phrased as a rule (a version pin, a lint config, a test threshold, a naming convention).
 3. **For each existing rule, decide:** is it a *project-wide non-negotiable* (→ ratify it as a principle, linking the stack article for detail) or a *local mechanic* (→ leave it in the stack wiki, do not lift it into the constitution)?
 4. **Contradictions are findings, not fixes.** If two stack articles disagree, or a stack article contradicts what the user states now, surface it and let the user resolve — never silently pick a winner.
@@ -137,12 +137,12 @@ The constitution is versioned so `analyze` and `review` can cite "constitution v
 - [ ] The Definition-of-Done checklist is present (what `verify` runs against).
 - [ ] Fixed principles included: human git authorship + decisions logged.
 - [ ] `02-DOCS/wiki/sdd/constitution.md` written with version + ratified date + amendment log.
-- [ ] Root `CLAUDE.md` Knowledge map has a row for the constitution (additive only — never delete sections).
+- [ ] Root `CLAUDE.md` `## Knowledge map` pointer has the read-first row for the constitution (it is one of the few entries kept in root `CLAUDE.md`; the full index lives in `02-DOCS/wiki/index.md`). Additive only — never delete sections.
 - [ ] The constitution was shown to the user and explicitly ratified.
 
 ## Project grounding (02-DOCS + CLAUDE.md)
 
-This skill's `02-DOCS` record is the constitution at `02-DOCS/wiki/sdd/constitution.md`, linked from a `## Knowledge map` row in the root `CLAUDE.md` (create `CLAUDE.md` if absent, additive only — never delete existing sections). Add this row if it is not already present:
+This skill's `02-DOCS` record is the constitution at `02-DOCS/wiki/sdd/constitution.md`. It is a **read-first** pointer entry, so its row stays in the short `## Knowledge map` pointer in the root `CLAUDE.md` (create `CLAUDE.md` if absent, additive only — never delete existing sections) — unlike other sdd artifacts, which are indexed in `02-DOCS/wiki/index.md` (the full Knowledge map that root `CLAUDE.md` points to). Add this row to the root pointer if it is not already present:
 
 ```markdown
 | Project constitution (SDD non-negotiables) | `02-DOCS/wiki/sdd/constitution.md` |
@@ -159,6 +159,6 @@ The constitution is the guardrail; now describe what to build. Hand off to **`..
 - `../sdd/SKILL.md` — the rsc-sdd dispatcher: the method, the phase map, the invoke rule.
 - `../specify/SKILL.md` — the next phase: intent → spec.
 - `../analyze/SKILL.md` — the consistency gate that checks specs/plans against this constitution.
-- `harness` — owns `02-DOCS/wiki/` (including `wiki/stack/*` and the Knowledge map this skill reconciles with and links into).
+- `harness` — owns `02-DOCS/wiki/` (including `wiki/stack/*` and the full Knowledge map at `02-DOCS/wiki/index.md` this skill reconciles with; the constitution's read-first row stays in the root `CLAUDE.md` pointer).
 - Stack skills the constitution *names but does not duplicate*: `../fastapi/SKILL.md`, `../nextjs/SKILL.md`, `../go/SKILL.md`, `../postgresdb/SKILL.md`, `../flutter/SKILL.md`, `../design/SKILL.md`, `../secure-coding/SKILL.md`.
 - References: `references/constitution-template.md`.

package/skills/continuous-learning/SKILL.md

@@ -92,7 +92,7 @@ Copy-pasteable and **situation-tagged**, so it retrieves on the trigger that mat
 - **We believed:** <the wrong assumption>
 - **Actually:** <what is true>
 - **Durable home:** <path + surface that now holds it>
-- **Fires next time via:** <rule / eval / verify.sh / Knowledge-map entry>
+- **Fires next time via:** <rule / eval / verify.sh / index entry in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer)>
 ```
 
 The last two lines are not optional. An entry with no durable home and no "fires next time" is a chat message with a date on it.
@@ -104,7 +104,7 @@ This is the rigor of this process skill, and it stands in for a `verify.sh` (thi
 - a rule the agent reads on its next pass (`CLAUDE.md` or a skill body), or
 - an eval case (`should_not_trigger` / `should_trigger`) that would have caught the miss, or
 - a `verify.sh` check on the owning skill that fails on the banned pattern, or
-- a `## Knowledge map` entry pointing at the new wiki article.
+- a `02-DOCS/wiki/index.md` entry (the Knowledge map; root `CLAUDE.md` keeps only a short pointer) pointing at the new wiki article.
 
 If you cannot name one of those, the lesson is not captured — keep going. Prefer a surface a human can review (a rule, an eval, a profile line) over an opaque store: a single agent reflecting alone can talk itself into a local optimum, so the durable write should be legible enough for a human to sanity-check.
 

package/skills/coolify/SKILL.md

@@ -186,14 +186,13 @@ live deploy. Exits 0 on a clean/empty target.
 ## Project grounding (02-DOCS + CLAUDE.md)
 
 When this skill runs in a project with a `02-DOCS/` layer (the harness Karpathy wiki), record this
-instance's deploy topology there and index it from the root `CLAUDE.md`, so the next agent inherits it
+instance's deploy topology there and index it in `02-DOCS/wiki/index.md`, so the next agent inherits it
 instead of re-deriving it.
 
-1. **Find the article** `02-DOCS/wiki/stack/coolify.md`, linked from a `## Knowledge map` section in the
-   root `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/coolify.md`, indexed in `02-DOCS/wiki/index.md` (the Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the real choices — the box (provider/specs), instance
    FQDN, which apps/databases run on it, build packs in use, the backup destination + cron + retention,
-   and the port/firewall decisions — then add/refresh the `CLAUDE.md` link.
+   and the port/firewall decisions — then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it).
 3. **Read it first on every use** and stay consistent; when the topology changes, update the article (bump
    its `Updated` date) in the same change. Never commit credentials here — record *where* secrets live,
    not their values.

package/skills/course-storytelling/SKILL.md

@@ -37,13 +37,13 @@ Do NOT use when (delegate or decline):
 
 Run this gate before reframing a single concept:
 
-1. **Locate the teaching profile.** Read the project's root `CLAUDE.md` and look for a `## Knowledge map` section linking into `02-DOCS/wiki/teaching/` (the `harness` Karpathy-wiki convention: compiled profile articles live under `02-DOCS/wiki/teaching/`, raw inputs the user pastes live under `02-DOCS/raw/teaching/`). If `CLAUDE.md` is absent, the link is missing, or it points nowhere, treat the profile as ABSENT.
+1. **Locate the teaching profile.** Read the project's root `CLAUDE.md` and follow its short `## Knowledge map` pointer to the full index at `02-DOCS/wiki/index.md`, looking for the entry into `02-DOCS/wiki/teaching/` (the `harness` Karpathy-wiki convention: compiled profile articles live under `02-DOCS/wiki/teaching/`, raw inputs the user pastes live under `02-DOCS/raw/teaching/`). If `CLAUDE.md` is absent, the index entry is missing, or it points nowhere, treat the profile as ABSENT.
 
 2. **Check completeness** against the checklist in `references/learner-grounding.md`. The profile is complete only when every dimension is filled: the LEARNER (level, prior knowledge, pains, desires, current false beliefs, what they want to DO after); the AUDIENCE (same as the buyer or not? live vs recorded? size? context?); the target TRANSFORMATION (the one result, before→after); and constraints/format. **Any empty dimension = INCOMPLETE.**
 
 3. **If ABSENT or INCOMPLETE, STOP and interview the user.** Ask the question script from `references/learner-grounding.md`, **one focused batch at a time** (do not dump all questions at once; ask, wait, persist, then continue). Then:
    - **a.** Write/update the profile into `02-DOCS/wiki/teaching/` as wiki articles (`learner.md`, `audience.md`, `transformation.md`, `false-beliefs.md`, `constraints.md`, plus an `index.md`), following the article format in `references/learner-grounding.md`. Save any raw material the user pastes (transcripts, outlines, existing slides) verbatim into `02-DOCS/raw/teaching/` and link it from the article's `> Raw:` line. Create the directories if they do not exist.
-   - **b.** Add or update a `## Knowledge map` section in the root `CLAUDE.md` linking to the teaching profile. Create `CLAUDE.md` if absent (additive only — never delete existing sections). The exact snippet is in `references/learner-grounding.md`.
+   - **b.** Index the teaching profile in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer). Ensure root `CLAUDE.md` carries that short pointer to the index, creating `CLAUDE.md` if absent (additive only — never delete existing sections). The exact snippet is in `references/learner-grounding.md`.
 
 4. **Only once the profile exists and is complete, proceed.** Cite which articles you used (e.g. "grounded in `02-DOCS/wiki/teaching/learner.md` and `false-beliefs.md`") so every reframing is traceable to a real learner, not an imagined one.
 
@@ -188,12 +188,12 @@ Automate → `scripts/verify.sh` (read-only; warns by default, `--strict` to gat
 
 ## Project grounding (02-DOCS + CLAUDE.md)
 
-This skill's 02-DOCS record has two parts, both indexed from a `## Knowledge map` section in the root `CLAUDE.md`:
+This skill's 02-DOCS record has two parts, both indexed in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer):
 
-- The **learner + audience profile** at `02-DOCS/wiki/teaching/` — a hard gate (see "Learner grounding" above): if the root `CLAUDE.md` lacks the link or any dimension is empty (learner, audience, transformation, false beliefs, constraints), ask until complete, persist it (raw inputs to `02-DOCS/raw/teaching/`), link it from `## Knowledge map` (create `CLAUDE.md` if absent), and only then teach.
+- The **learner + audience profile** at `02-DOCS/wiki/teaching/` — a hard gate (see "Learner grounding" above): if the index lacks the entry or any dimension is empty (learner, audience, transformation, false beliefs, constraints), ask until complete, persist it (raw inputs to `02-DOCS/raw/teaching/`), index it in `02-DOCS/wiki/index.md` and ensure root `CLAUDE.md` carries the short pointer to that index (create `CLAUDE.md` if absent), and only then teach.
 - The **course teaching conventions** at `02-DOCS/wiki/stack/course-storytelling.md` (or alongside the profile under `02-DOCS/wiki/teaching/`) — the established narrative spine, the named mental models already coined, the Big Dominoes per module, and the teacher's Attractive Character. Recorded, not gated.
 
-Create/update both as decisions are made and refresh their `CLAUDE.md` links. Read them first on every use and keep every reframing consistent with them. If the project has no `02-DOCS` layer at all, skip this section silently and proceed with the in-session profile.
+Create/update both as decisions are made and keep their entries current in `02-DOCS/wiki/index.md` (root `CLAUDE.md` keeps only a short pointer to it). Read them first on every use and keep every reframing consistent with them. If the project has no `02-DOCS` layer at all, skip this section silently and proceed with the in-session profile.
 
 ## See Also
 

package/skills/cpp/SKILL.md

@@ -314,15 +314,13 @@ bad index is UB, not an exception.
 ## Project grounding (02-DOCS + CLAUDE.md)
 
 When this skill runs in a project with a `02-DOCS/` layer (the [`harness`](../harness/SKILL.md)
-Karpathy wiki), record this project's C++ decisions there and index them from the root `CLAUDE.md`,
+Karpathy wiki), record this project's C++ decisions there and index them in `02-DOCS/wiki/index.md`,
 so the next agent inherits them instead of re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/cpp.md`, linked from a `## Knowledge map` section in
-   the root `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/cpp.md`, indexed in `02-DOCS/wiki/index.md` (the Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — std version and
    compiler matrix, CMake layout and presets, the sanitizer/warning policy, and the ownership/error
-   conventions — then add/refresh the `CLAUDE.md` link (create the `## Knowledge map` section, and
-   `CLAUDE.md` itself, if absent).
+   conventions — then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the article
    (bump its `Updated` date) in the same change.
 

package/skills/decision-records/SKILL.md

@@ -40,7 +40,7 @@ Prefer **numeric-prefix**: `NNNN-title-with-dashes.md`. Stable IDs are what make
 **Where to store** (first match wins):
 1. The project's existing convention — never invent a parallel log.
 2. A code repo with no convention: `docs/adr/` (also seen as `doc/adr/`).
-3. A harness workspace: `02-DOCS/wiki/decisions/`, linked from the wiki knowledge map.
+3. A harness workspace: `02-DOCS/wiki/decisions/`, indexed in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer).
 
 ```text
 Bad:  decision-final-v2-REAL.md          # no ID, no order, will rot
@@ -169,7 +169,7 @@ A decision log is only useful if it stays navigable.
 - **Index every ADR at creation.** One row: `| 0007 | Choose Postgres | accepted | 2026-06-02 |`. No orphans.
 - **Keep supersession links live** in both directions.
 - **Review cadence.** Periodically sweep `accepted` ADRs — anything reality has overtaken gets a superseding record, not a silent edit.
-- **Link the log from the entry point** — root `CLAUDE.md` or the wiki knowledge map — so onboarding finds it. The broader wiki and onboarding doc are owned by `knowledge-ops` and `codebase-onboarding`; this skill owns only the ADRs the wiki links to. The meeting that spawned a decision routes to [`meeting-notes`](../meeting-notes/SKILL.md); the repeatable how-to it implies routes to [`sop-builder`](../sop-builder/SKILL.md).
+- **Index the log from the entry point** — `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer) — so onboarding finds it. The broader wiki and onboarding doc are owned by `knowledge-ops` and `codebase-onboarding`; this skill owns only the ADRs the wiki links to. The meeting that spawned a decision routes to [`meeting-notes`](../meeting-notes/SKILL.md); the repeatable how-to it implies routes to [`sop-builder`](../sop-builder/SKILL.md).
 
 ## Anti-patterns
 

package/skills/deployment/SKILL.md

@@ -409,14 +409,12 @@ async def readyz() -> dict[str, str]:
 
 When this skill runs in a project with a `02-DOCS/` layer (the
 [`harness`](../harness/SKILL.md) Karpathy wiki), record this
-project's deploy decisions there and index them from the root `CLAUDE.md`, so the next
+project's deploy decisions there and index them in `02-DOCS/wiki/index.md`, so the next
 agent inherits the conventions instead of re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/deployment.md`, linked from a `## Knowledge map` section in the root
-   `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/deployment.md`, indexed in `02-DOCS/wiki/index.md` (the Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — the base-image/container choices, the CI pipeline, the Coolify/target config, the secrets flow, and the rollback strategy —
-   then add/refresh the `CLAUDE.md` link (create the `## Knowledge map` section, and
-   `CLAUDE.md` itself, if absent).
+   then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the
    article (bump its `Updated` date) in the same change.
 

package/skills/design/SKILL.md

@@ -368,8 +368,8 @@ Score = Σ(dimension × weight), max 115 (normalize to /100 by ×100/115 if you
 
 ## Project grounding (02-DOCS + CLAUDE.md)
 
-This skill's 02-DOCS record has two parts, both indexed from a `## Knowledge map` section in
-the root `CLAUDE.md`:
+This skill's 02-DOCS record has two parts, both indexed in `02-DOCS/wiki/index.md` (the Knowledge
+map index; root `CLAUDE.md` points to it):
 
 - The **brand study** at `02-DOCS/wiki/brand/` — a hard gate (see "Brand grounding" above): if
   missing or incomplete, ask until complete before designing.
@@ -377,8 +377,8 @@ the root `CLAUDE.md`:
   (color/OKLCH, type scale, spacing, radius, shadow, motion), the 2026 direction picked, and the
   reference sites. Recorded, not gated.
 
-Create/update both as decisions are made and add/refresh their `CLAUDE.md` links (create the
-`## Knowledge map` section, and `CLAUDE.md` itself, if absent). Read them first on every use and
+Create/update both as decisions are made and index them in `02-DOCS/wiki/index.md` (the Knowledge map;
+root `CLAUDE.md` keeps only a short pointer to it). Read them first on every use and
 keep outputs consistent with them.
 
 ## See Also

package/skills/duckdb/SKILL.md

@@ -203,5 +203,6 @@ version. If neither the CLI nor the Python module is present it prints `SKIP` an
 
 In a project with a `02-DOCS/` layer (the [`harness`](../harness/SKILL.md) wiki), record this project's
 DuckDB decisions — version pin, file layout, persistent vs in-memory, remote/secret setup — in
-`02-DOCS/wiki/stack/duckdb.md` and link it from the root `CLAUDE.md` `## Knowledge map`. Read it first on
+`02-DOCS/wiki/stack/duckdb.md` and index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md`
+keeps only a short pointer to it). Read it first on
 every use and keep choices consistent. No `02-DOCS/`? Skip silently. Conventions are recorded, never gated.

package/skills/fastapi/SKILL.md

@@ -480,14 +480,14 @@ shutdown, keyset pagination, caching, `ORJSONResponse` and proxy headers all liv
 
 When this skill runs in a project with a `02-DOCS/` layer (the
 [`harness`](../harness/SKILL.md) Karpathy wiki), record this project's API decisions in
-`02-DOCS/wiki/stack/fastapi.md` and link it from a `## Knowledge map` section in the root
-`CLAUDE.md`, so the next agent inherits the conventions instead of re-deriving them.
+`02-DOCS/wiki/stack/fastapi.md` and index it in `02-DOCS/wiki/index.md` (the Knowledge map index; root
+`CLAUDE.md` points to it), so the next agent inherits the conventions instead of re-deriving them.
 
 - **Read it first** on every use and stay consistent; bump its `Updated` date when a
   convention changes.
 - **Create/update it** with the project's real choices — auth model (JWT/OAuth2 provider,
   token TTLs), DB session + migration tool, error-envelope shape, settings/secrets approach,
-  deployment target — adding the `CLAUDE.md` link (and the file) if absent.
+  deployment target — indexing it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it) if absent.
 
 No `02-DOCS/` layer? Skip silently (optionally suggest `harness`). Technical conventions are
 *recorded, not gated* — never block the task on this.

package/skills/flutter/SKILL.md

@@ -423,14 +423,12 @@ encapsulation → `references/i18n-and-dependencies.md`.
 
 When this skill runs in a project with a `02-DOCS/` layer (the
 [`harness`](../harness/SKILL.md) Karpathy wiki), record this
-project's app decisions there and index them from the root `CLAUDE.md`, so the next
+project's app decisions there and index them in `02-DOCS/wiki/index.md`, so the next
 agent inherits the conventions instead of re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/flutter.md`, linked from a `## Knowledge map` section in the root
-   `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/flutter.md`, indexed in `02-DOCS/wiki/index.md` (the Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — the state-management choice (Riverpod/Bloc), the architecture layers, routing, the Material 3 token system, and codegen setup —
-   then add/refresh the `CLAUDE.md` link (create the `## Knowledge map` section, and
-   `CLAUDE.md` itself, if absent).
+   then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the
    article (bump its `Updated` date) in the same change.
 

package/skills/go/SKILL.md

@@ -411,14 +411,12 @@ Docker note: distroless/static base, `CGO_ENABLED=0`, multi-stage build. Full Co
 
 When this skill runs in a project with a `02-DOCS/` layer (the
 [`harness`](../harness/SKILL.md) Karpathy wiki), record this
-project's service decisions there and index them from the root `CLAUDE.md`, so the next
+project's service decisions there and index them in `02-DOCS/wiki/index.md`, so the next
 agent inherits the conventions instead of re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/go.md`, linked from a `## Knowledge map` section in the root
-   `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/go.md`, indexed in `02-DOCS/wiki/index.md` (the Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — the project layout, the router (stdlib 1.22 / chi), the error and `slog` logging conventions, and concurrency/timeout defaults —
-   then add/refresh the `CLAUDE.md` link (create the `## Knowledge map` section, and
-   `CLAUDE.md` itself, if absent).
+   then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the
    article (bump its `Updated` date) in the same change.
 

package/skills/harness/SKILL.md

@@ -276,6 +276,31 @@ Once the structure stands, make sure the workspace has the rsc skills its stack
    Can't run a shell? Print the exact command for another terminal tab.
 3. **Flag the new session.** New skills load at session start — tell the user to open a **new tab/session** (or reload Cursor/Codex/Gemini) in this folder for them to activate. Log the installed set in `02-DOCS/wiki/harness/decisions.md`.
 
+## Keep CLAUDE.md lean — the index lives in the wiki
+
+Root `CLAUDE.md` is read on **every** turn, so every line is a permanent context tax (2026 best
+practice: keep it **under ~200 lines**; beyond that, adherence rots as the rules that matter get
+diluted by an index nobody needs in context). The biggest growth vector is the `## Knowledge map` —
+a row per wiki article, appended by many skills, forever.
+
+**The rule:** the **full** Knowledge map lives in `02-DOCS/wiki/index.md` (loaded on demand, grows
+freely). Root `CLAUDE.md`'s `## Knowledge map` is a **short pointer** — only the read-first entries
+(`harness/user-profile.md`, `sdd/constitution.md`) plus "full index → `02-DOCS/wiki/index.md`".
+
+**Offload when it bloats (a move, never a trim — no info lost):** when `CLAUDE.md` passes ~200 lines
+(the SessionStart hook nudges you) or its `## Knowledge map` has grown past the read-first entries:
+
+1. Open `02-DOCS/wiki/index.md` (create it if absent).
+2. **Move** every Knowledge-map row beyond the read-first entries from `CLAUDE.md` into
+   `02-DOCS/wiki/index.md`, merging — don't duplicate, don't delete.
+3. Leave `CLAUDE.md`'s `## Knowledge map` as the short pointer above.
+4. Same for any other section overgrown into an index (e.g. a huge tool table): detail to the wiki,
+   pointer stays.
+
+From then on, **new index entries go to `02-DOCS/wiki/index.md`**, not `CLAUDE.md`. This is additive
+and reversible; it honors the "never delete user content" rule (you relocate it, with a pointer).
+Opt out of the size nudge with `.rsc/.no-claudemd-check`.
+
 ## Iron rules (non-negotiable)
 
 These rules cut across every phase. Violating any one of them aborts the run.

package/skills/harness/references/claude-md-template.md

@@ -14,32 +14,19 @@ workspace. If you're working in a subproject that also has its own
 
 ## Knowledge map
 
-The living record of this project's conventions lives under `02-DOCS/wiki/` and is maintained by
-the stack skills, plus the `harness/` topic maintained by the harness itself (user profile +
-decisions log). Each skill links its article here as it is created — read the relevant article
-before working in its area (the `harness/user-profile.md` first, to match the user's technical +
-accompaniment level), and update it when conventions change.
+The **full** index of this project's conventions lives in **`02-DOCS/wiki/index.md`** — maintained by
+the stack skills plus the `harness/` topic. Read the relevant article before working in its area, and
+update it when conventions change. Keep this section a **short pointer**: CLAUDE.md is read on every
+turn, so the full per-area table belongs in `02-DOCS/wiki/index.md`, not here. New index entries go
+there, not into this file.
+
+Read first, always:
 
 | Area | Article |
 |------|---------|
 | User profile (technical + accompaniment level) | `02-DOCS/wiki/harness/user-profile.md` |
-| Decisions log (append-only, "siempre 3 opciones") | `02-DOCS/wiki/harness/decisions.md` |
 | SDD constitution (project non-negotiables) | `02-DOCS/wiki/sdd/constitution.md` |
-| SDD specs (what & why, per feature) | `02-DOCS/wiki/sdd/specs/` |
-| SDD plans (technical approach, per feature) | `02-DOCS/wiki/sdd/plans/` |
-| SDD decisions (implement-phase log) | `02-DOCS/wiki/sdd/decisions.md` |
-| Brand & voice | `02-DOCS/wiki/brand/` |
-| Design system | `02-DOCS/wiki/stack/design.md` |
-| FastAPI / API | `02-DOCS/wiki/stack/fastapi.md` |
-| Next.js app | `02-DOCS/wiki/stack/nextjs.md` |
-| Go services | `02-DOCS/wiki/stack/go.md` |
-| PostgreSQL | `02-DOCS/wiki/stack/postgresdb.md` |
-| Flutter app | `02-DOCS/wiki/stack/flutter.md` |
-| Agents | `02-DOCS/wiki/stack/agents.md` |
-| Security | `02-DOCS/wiki/stack/security.md` |
-| Deployment | `02-DOCS/wiki/stack/deployment.md` |
-| Presentations | `02-DOCS/wiki/stack/presentations.md` |
-| Course storytelling | `02-DOCS/wiki/stack/course-storytelling.md` |
+| **Everything else — full index** | `02-DOCS/wiki/index.md` |
 
 ## Workspace map
 

package/skills/implement/SKILL.md

@@ -158,6 +158,12 @@ test output. You merge the branches, then run the *combined* test suite before t
 green-in-isolation is not green-together. Hand the orchestration to `parallel`; keep the TDD
 discipline inside each branch.
 
+**Dispatch implementation work to the `developer` subagent.** rsc installs a `developer` agent
+pinned to the balanced tier (Sonnet by default, chosen at onboarding, never `light`). When you
+delegate a task or fan out via `parallel`, dispatch it to that agent (e.g. Claude Code
+`subagent_type: developer`) so the bulk of TDD execution runs on the cheaper-but-capable model
+while you stay on the session model to orchestrate. Full rationale: `../sdd/references/model-routing.md`.
+
 ## Model tier — `balanced` (opt-in routing)
 
 This phase's default model tier is **`balanced`** — it is the bulk of TDD execution: cost-sensitive, with quality balanced handles well. Routing is **off** unless `models.enabled: true` in `02-DOCS/wiki/sdd/config.yaml`. When on: resolve this phase's tier (`models.overrides` wins over `models.phases`), map it to a model via `models.tiers`, and apply per `../sdd/references/model-routing.md` — announce the switch per the accompaniment dial when it differs from the session model, and dispatch any `Task`/`parallel` subagents on that model (this is where routing pays off most — fan-out runs on `balanced` while a hard sub-problem can be escalated to `heavy`). Routing off or no profile → honor the session model silently. Never fake a switch a tool can't make; skip routing on a one-line change.
@@ -180,7 +186,7 @@ done-checks, the constitution and decision logging hold at every level, includin
 
 When you make a choice the plan did not fully specify — a library, a data shape, an error contract, a
 deviation from the plan — append it to `02-DOCS/wiki/sdd/decisions.md` (append-only; create it if
-absent and add a row to the root `CLAUDE.md` `## Knowledge map` under the `sdd/` topic). One entry:
+absent and index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer) under the `sdd/` topic). One entry:
 
 ```text
 ## YYYY-MM-DD — <short title>  (feature: <slug>, task: <n>)

package/skills/init/SKILL.md

@@ -51,10 +51,18 @@ Before discovery, before any recommendation, write the profile to `02-DOCS` and
 
 - `02-DOCS/wiki/harness/user-profile.md` — the living profile (technical level, accompaniment level, goals, context, constraints). Format in `references/accompaniment-and-profile.md`.
 - `02-DOCS/wiki/harness/decisions.md` — an **append-only** decisions log. Every significant decision gets one entry, never edited or deleted. Format in `references/accompaniment-and-profile.md`.
-- Root `CLAUDE.md` → a `## Knowledge map` section linking to BOTH files. Create `CLAUDE.md` if absent (additive only — never delete existing sections).
+- Root `CLAUDE.md` → a SHORT `## Knowledge map` pointer: the read-first entries (this `user-profile.md` and the `decisions.md` log) plus a "full index → `02-DOCS/wiki/index.md`" line. Keep this pointer tiny — it loads on every turn; ALL other index entries live in `02-DOCS/wiki/index.md`, not here. Create `CLAUDE.md` if absent (additive only — never delete existing sections).
 
 If `02-DOCS/` does not yet exist (greenfield), create `02-DOCS/wiki/harness/` now — just enough to hold these two files. That, plus the `CLAUDE.md` Knowledge-map link, is everything `init` writes; ALL other `01-TOOLS/` + `02-DOCS/` scaffolding is the `harness` skill's job.
 
+### Step 4 — Propose the developer model (the implementation subagent)
+
+rsc installs a **`developer`** subagent (the fan-out / implementation worker) for every assistant that supports file-based agents (Claude Code, Cursor, OpenCode, Gemini, Copilot, Junie, Kiro, Codex). It runs at the **balanced** tier by default — **Sonnet** on Anthropic tools (the provider's mid model elsewhere) — and **never** the cheapest `light` model, which is too weak to build with. Offer the choice once, calibrated to the dial:
+
+- *"La implementación la hará un sub-agente `developer`. ¿Qué modelo? **balanced / Sonnet** (rápido y económico — recomendado) o **heavy / Opus** (máxima calidad, más caro)."* Never offer `light`/Haiku.
+
+Record it to `.rsc/developer.json` (`{ "tier": "balanced" }` or `"heavy"`) and re-run the install/sync so the agent files adopt it. If you skip the question (e.g. L0), default to `balanced`. The default is also written at install time, so the developer agent is always set even when onboarding never asks — "propose at onboarding, and also when it isn't set."
+
 ### Opt-out marker — `.rsc/.no-harness`
 
 A freshly-installed session auto-starts `init` while `02-DOCS/wiki/harness/user-profile.md` is absent (the `suggest` Onboarding gate + claude's SessionStart hook). If the user does not want a harness in this repo (e.g. they installed only code skills), write an empty `.rsc/.no-harness` — this permanently silences the auto-start here even before a profile exists. Completing first contact (which writes `user-profile.md`) also silences it. The marker is project-local; commit it so the "no harness here" decision is shared by the team.
@@ -196,7 +204,7 @@ These thoughts mean the skill is about to break its own rules. Recognize and abo
 
 ## Project grounding (02-DOCS + CLAUDE.md)
 
-This skill's `02-DOCS` record is the **user profile** at `02-DOCS/wiki/harness/user-profile.md` plus the append-only **decisions log** at `02-DOCS/wiki/harness/decisions.md`. Both are written in Phase 1 and updated throughout, and both are linked from a `## Knowledge map` section in the root `CLAUDE.md` (created if absent, additive only). Every rsc skill reads the profile first and adapts its verbosity and question count to `accompaniment_level` and `technical_level`. Those two files plus the Knowledge-map link are everything `init` writes; ALL other `01-TOOLS/` + `02-DOCS/` scaffolding is the `harness` skill's job, and it reads this same profile.
+This skill's `02-DOCS` record is the **user profile** at `02-DOCS/wiki/harness/user-profile.md` plus the append-only **decisions log** at `02-DOCS/wiki/harness/decisions.md`. Both are written in Phase 1 and updated throughout. Because they are the read-first entries, both stay in the SHORT `## Knowledge map` pointer in the root `CLAUDE.md` (created if absent, additive only) — alongside a "full index → `02-DOCS/wiki/index.md`" line where every OTHER index entry lives. Every rsc skill reads the profile first and adapts its verbosity and question count to `accompaniment_level` and `technical_level`. Those two files plus the Knowledge-map link are everything `init` writes; ALL other `01-TOOLS/` + `02-DOCS/` scaffolding is the `harness` skill's job, and it reads this same profile.
 
 Verify the profile and the Knowledge-map link exist with `scripts/verify.sh` (read-only; warns, never fails).
 

package/skills/java/SKILL.md

@@ -303,13 +303,13 @@ When this skill runs in a project with a `02-DOCS/` layer (the
 [`harness`](../harness/SKILL.md) Karpathy wiki), record this project's Java decisions there
 and index them from the root `CLAUDE.md`, so the next agent inherits them instead of re-deriving.
 
-1. **Find the article** `02-DOCS/wiki/stack/java.md`, linked from a `## Knowledge map` section
-   in the root `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/java.md`, indexed in `02-DOCS/wiki/index.md` (the
+   Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — JDK/LTS target,
    Maven vs Gradle, the domain-modeling conventions (records/sealed), the concurrency model
    (virtual threads, structured-concurrency preview on/off), and the error/null conventions —
-   then add/refresh the `CLAUDE.md` link (create the `## Knowledge map` section and `CLAUDE.md`
-   itself if absent).
+   then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a
+   short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the
    article (bump its `Updated` date) in the same change.
 

package/skills/marketing/SKILL.md

@@ -221,9 +221,9 @@ Automate → `scripts/verify.sh` (read-only grep gate; warns by default, `--stri
 This skill's 02-DOCS record is the **brand & voice study** at `02-DOCS/wiki/brand/` — a hard
 gate (see "Brand grounding" above): if the root `CLAUDE.md` lacks the link or the study is
 incomplete (no voice samples, positioning, do/don't lists), ask until complete, persist it, and
-link it from a `## Knowledge map` section in `CLAUDE.md` (create `CLAUDE.md` if absent). Read it
-first on every use and ground all copy in it. Site and visual conventions belong to the sibling
-`design` and `nextjs` articles in the same Knowledge map.
+index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short
+pointer to it). Read it first on every use and ground all copy in it. Site and visual conventions
+belong to the sibling `design` and `nextjs` articles in the same Knowledge map index.
 
 ## See Also
 

package/skills/modal/SKILL.md

@@ -296,7 +296,8 @@ exits 0.
 
 In a project with a `02-DOCS/` layer (the [`harness`](../harness/SKILL.md) wiki), record this
 app's real Modal choices — GPU types, image base, Volume names, schedule, endpoint shape — in
-`02-DOCS/wiki/stack/modal.md` and link it from the root `CLAUDE.md` `## Knowledge map`. Read it
+`02-DOCS/wiki/stack/modal.md` and index it in `02-DOCS/wiki/index.md` (the Knowledge map; root
+`CLAUDE.md` keeps only a short pointer to it). Read it
 first on every use; create/update it with the real decisions. No `02-DOCS/`? Skip silently.
 
 ## See Also

package/skills/mongodb/SKILL.md

@@ -311,12 +311,12 @@ When this skill runs in a project with a `02-DOCS/` layer (the [`harness`](../ha
 Karpathy wiki), record this project's MongoDB decisions there and index them from the root
 `CLAUDE.md`, so the next agent inherits the conventions instead of re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/mongodb.md`, linked from a `## Knowledge map` section in
-   the root `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/mongodb.md`, indexed in `02-DOCS/wiki/index.md` (the
+   Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — collection layout and
    embed/reference decisions, the index set and its ESR rationale, read/write concern policy, the
-   Atlas tier, and any encryption/RBAC setup — then add/refresh the `CLAUDE.md` link (create the
-   `## Knowledge map` section, and `CLAUDE.md` itself, if absent).
+   Atlas tier, and any encryption/RBAC setup — then index it in `02-DOCS/wiki/index.md` (the
+   Knowledge map; root `CLAUDE.md` keeps only a short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the article
    (bump its `Updated` date) in the same change.
 

package/skills/nextjs/SKILL.md

@@ -454,11 +454,11 @@ When this skill runs in a project with a `02-DOCS/` layer (the
 project's app decisions there and index them from the root `CLAUDE.md`, so the next
 agent inherits the conventions instead of re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/nextjs.md`, linked from a `## Knowledge map` section in the root
-   `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/nextjs.md`, indexed in `02-DOCS/wiki/index.md` (the
+   Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — the caching model in use (v15 fetch-cache vs v16 `use cache`), the auth approach, server-action and data-fetching conventions, runtime choices (edge/node), and the design-system hookup —
-   then add/refresh the `CLAUDE.md` link (create the `## Knowledge map` section, and
-   `CLAUDE.md` itself, if absent).
+   then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a
+   short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the
    article (bump its `Updated` date) in the same change.
 

package/skills/parallel/SKILL.md

@@ -75,6 +75,8 @@ Each subagent still owns its own discipline inside its scope — TDD via `implem
 
 **Per-unit model tier (when routing is enabled).** `parallel` has *no fixed tier* — this is the most concrete place per-phase model routing pays off. When `models.enabled: true` in `02-DOCS/wiki/sdd/config.yaml`, give each unit the tier of the *kind of work it does*, not one tier for the whole fan-out: an implement-type unit → `balanced`, a scan/research/boilerplate unit → `light`, a unit doing genuine design or root-cause reasoning → `heavy`. Resolve the tier to a concrete model via `models.tiers` and **dispatch that subagent on that model** (e.g. Claude Code's `model` field on the Task/subagent) — real routing, independent of the session model. If routing is off or no profile exists, dispatch on the session model and say nothing. Full protocol: `../sdd/references/model-routing.md`.
 
+**The `developer` agent is the default worker.** rsc installs a `developer` subagent pinned to the **balanced** tier (Sonnet by default; the user's onboarding choice in `.rsc/developer.json`, never `light`). For implement-type units, **dispatch to the `developer` agent** (e.g. Claude Code `subagent_type: developer`) — that's the deliberate cost cap the user asked for. Only escalate a genuinely heavy unit (real design / root-cause) to a heavy model, and only when routing is enabled; otherwise the `developer` agent's balanced model is the floor and the ceiling.
+
 ### Skill resolution feedback
 
 Every subagent result must report:

package/skills/plan/SKILL.md

@@ -51,7 +51,7 @@ Run this gate. Skipping it produces a plan that drifts from intent on contact.
 3. **Read the constitution.** `02-DOCS/wiki/sdd/constitution.md` holds the project's non-negotiables
    (stack canon, quality bars, conventions). Every architectural choice must be consistent with it;
    when the plan needs to bend a principle, say so explicitly with a reason — don't bend it silently.
-4. **Read the Knowledge map.** The root `CLAUDE.md` `## Knowledge map` points at existing
+4. **Read the Knowledge map.** The full index at `02-DOCS/wiki/index.md` (root `CLAUDE.md` keeps only a short pointer to it) points at existing
    `02-DOCS/wiki/stack/*` and prior plans/decisions. Reuse what's already established. A plan that
    reinvents a pattern the project already settled is scope drift.
 
@@ -172,8 +172,8 @@ complete; it's just quiet.
 
 Write to `02-DOCS/wiki/sdd/plans/<slug>.md`, where `<slug>` matches the spec's slug exactly (one
 plan per spec, same name — that's how `tasks`, `analyze`, and `implement` find it). Use
-`references/plan-template.md` verbatim as the skeleton. Then index it: add a row under the
-`## Knowledge map` section of the root `CLAUDE.md` pointing at the new plan, so the harness wiki and
+`references/plan-template.md` verbatim as the skeleton. Then index it in `02-DOCS/wiki/index.md`
+(the Knowledge map; root `CLAUDE.md` keeps only a short pointer) pointing at the new plan, so the harness wiki and
 every later phase can find it. If a plan for this slug already exists, update it in place and note
 what changed — don't fork a `-v2`.
 

package/skills/postgresdb/SKILL.md

@@ -353,11 +353,11 @@ When this skill runs in a project with a `02-DOCS/` layer (the
 project's database decisions there and index them from the root `CLAUDE.md`, so the next
 agent inherits the conventions instead of re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/postgresdb.md`, linked from a `## Knowledge map` section in the root
-   `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/postgresdb.md`, indexed in `02-DOCS/wiki/index.md` (the
+   Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — schema and naming conventions, the migration tool, indexing/partitioning decisions, the pooling setup, and any RLS policies —
-   then add/refresh the `CLAUDE.md` link (create the `## Knowledge map` section, and
-   `CLAUDE.md` itself, if absent).
+   then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a
+   short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the
    article (bump its `Updated` date) in the same change.
 

package/skills/presentations/SKILL.md

@@ -276,8 +276,8 @@ When this skill runs in a project with a `02-DOCS/` layer (the
 conventions there and index them from the root `CLAUDE.md`, so the next agent inherits them instead of
 re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/presentations.md`, linked from a `## Knowledge map` section in the root `CLAUDE.md`.
-2. **If missing or stale**, create/update it with this project's real choices — the chosen pipeline (Marp / Slidev / python-pptx) and why; the theme file path and how it maps the design tokens (`02-DOCS/wiki/stack/design.md`); the standard deck arc(s); export commands and the canonical output (PDF / PPTX); presenter-vs-leave-behind convention; font-embedding and asset-location notes — then add/refresh the `CLAUDE.md` link (create the `## Knowledge map` section, and `CLAUDE.md` itself, if absent).
+1. **Find the article** `02-DOCS/wiki/stack/presentations.md`, indexed in `02-DOCS/wiki/index.md` (the Knowledge map index; root `CLAUDE.md` points to it).
+2. **If missing or stale**, create/update it with this project's real choices — the chosen pipeline (Marp / Slidev / python-pptx) and why; the theme file path and how it maps the design tokens (`02-DOCS/wiki/stack/design.md`); the standard deck arc(s); export commands and the canonical output (PDF / PPTX); presenter-vs-leave-behind convention; font-embedding and asset-location notes — then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the article (bump its `Updated` date) in the same change.
 
 The deck theme is downstream of the design tokens: always reconcile `02-DOCS/wiki/stack/presentations.md`

package/skills/python/SKILL.md

@@ -348,9 +348,9 @@ When this skill runs in a project with a `02-DOCS/` layer (the
 [`harness`](../harness/SKILL.md) Karpathy wiki), record this project's Python conventions
 there so the next agent inherits them — *recorded, not gated*, never block the task on this.
 
-1. **Find** `02-DOCS/wiki/stack/python.md`, linked from a `## Knowledge map` in the root `CLAUDE.md`.
+1. **Find** `02-DOCS/wiki/stack/python.md`, indexed in `02-DOCS/wiki/index.md` (the Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, write the project's real choices (interpreter floor, `src/` layout,
-   uv workflow, ruff/mypy config, async-vs-sync stance, data-modeling defaults) and add the link.
+   uv workflow, ruff/mypy config, async-vs-sync stance, data-modeling defaults) and index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it).
 3. **Read it first on every use**; bump its `Updated` date when a convention changes.
 
 No `02-DOCS/` layer? Skip silently (optionally suggest `harness`).

package/skills/review/SKILL.md

@@ -219,7 +219,7 @@ Review is mostly a conversation, but two artifacts persist into the harness wiki
 - **Accepted/declined findings** of consequence → append to `02-DOCS/wiki/sdd/decisions.md` (e.g. "declined NPE finding on documents.py:42 — guarded by require_auth; added regression test"). This is the same append-only log `implement` writes to.
 - If a finding reveals a **missing constitution rule** (a defect class the project keeps hitting), propose adding it to `02-DOCS/wiki/sdd/constitution.md` so the next review catches it earlier.
 
-Index both under the root `CLAUDE.md` `## Knowledge map` (the `sdd/` topic) — the harness owns that map; this skill just keeps its rows honest.
+Index both in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer) under the `sdd/` topic — the harness owns that map; this skill just keeps its rows honest.
 
 ## Next in the chain
 

package/skills/rust/SKILL.md

@@ -365,12 +365,12 @@ When this skill runs in a project with a `02-DOCS/` layer (the
 [`harness`](../harness/SKILL.md) Karpathy wiki), record this project's service decisions there and index
 them from the root `CLAUDE.md`, so the next agent inherits the conventions instead of re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/rust.md`, linked from a `## Knowledge map` section in the
-   root `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/rust.md`, indexed in `02-DOCS/wiki/index.md` (the
+   Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — the crate/workspace layout,
    the runtime (tokio), the HTTP framework (axum 0.8), the error strategy (thiserror enum + IntoResponse
-   mapping), the DB layer (sqlx + pool), and tracing/concurrency defaults — then add/refresh the
-   `CLAUDE.md` link (create the `## Knowledge map` section, and `CLAUDE.md` itself, if absent).
+   mapping), the DB layer (sqlx + pool), and tracing/concurrency defaults — then index it in
+   `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the article
    (bump its `Updated` date) in the same change.
 

package/skills/sdd/SKILL.md

@@ -145,9 +145,10 @@ Every phase writes under `02-DOCS/wiki/sdd/` so the feature's reasoning outlives
 └── decisions.md             ← append-only log of decisions taken while building
 ```
 
-Index these from the root `CLAUDE.md` `## Knowledge map` under an `sdd/` topic, so every other skill reads them before working in the area. The harness maintains and improves these files just like any other wiki topic — `sdd` produces them, the harness keeps them honest.
+Index these in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer) under an `sdd/` topic, so every other skill reads them before working in the area. The harness maintains and improves these files just like any other wiki topic — `sdd` produces them, the harness keeps them honest.
 
 ```markdown
+# 02-DOCS/wiki/index.md
 ## Knowledge map
 | Topic | Where | What |
 | --- | --- | --- |

package/skills/sdd/references/model-routing.md

@@ -155,6 +155,24 @@ To target another provider, change `models.provider` and set `models.tiers` to t
 These are starting points — set them to whatever your account actually has. The tiers are the
 contract; the concrete names are yours to edit.
 
+## The `developer` subagent (the installed implementation worker)
+
+rsc installs a **`developer`** subagent into every assistant that supports file-based agents
+(Claude Code `.claude/agents/`, Cursor, OpenCode, Gemini, Copilot, Junie, Kiro, Codex) when the
+harness is installed. It is the concrete embodiment of "run the fan-out on a smaller model":
+
+- It is pinned to the **balanced** tier — **Sonnet** for Anthropic-backed tools, the provider's
+  mid model elsewhere (Gemini Flash, GPT-mini). It is **never** `light`/Haiku — that tier is too
+  weak to build with. The floor for implementation is balanced.
+- The tier is chosen at **onboarding** (`init`, Step 4): balanced (default) or heavy. It is stored
+  in `.rsc/developer.json` and the installer writes the per-target agent file with the resolved
+  model on every (re)install/sync. If onboarding never asks, the install-time default is balanced.
+- `implement` and `parallel` **dispatch implementation/fan-out work to this agent**, so the
+  session model orchestrates while the cheaper-but-capable model does the bulk of the typing.
+- On assistants without file-based agents (Amp, Zed, Windsurf, Cline, Roo, Continue, Aider, Jules,
+  Antigravity), there is no agent file — the developer tier is advisory there, like the rest of
+  routing. The per-target model id is an editable default (see the provider table above).
+
 ## Result-envelope `model` field
 
 Every phase's result envelope carries the model it ran on, so the chain is auditable:

package/skills/secure-coding/SKILL.md

@@ -336,11 +336,11 @@ When this skill runs in a project with a `02-DOCS/` layer (the
 project's security decisions there and index them from the root `CLAUDE.md`, so the next
 agent inherits the conventions instead of re-deriving them.
 
-1. **Find the article** `02-DOCS/wiki/stack/security.md`, linked from a `## Knowledge map` section in the root
-   `CLAUDE.md`.
+1. **Find the article** `02-DOCS/wiki/stack/security.md`, indexed in `02-DOCS/wiki/index.md` (the
+   Knowledge map index; root `CLAUDE.md` points to it).
 2. **If missing or stale**, create/update it with the project's real choices — the threat model, the auth model, the secrets backend, the CI security gates, and any accepted risks —
-   then add/refresh the `CLAUDE.md` link (create the `## Knowledge map` section, and
-   `CLAUDE.md` itself, if absent).
+   then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a
+   short pointer to it).
 3. **Read it first on every use** and stay consistent; when a convention changes, update the
    article (bump its `Updated` date) in the same change.
 

package/skills/security-scan/SKILL.md

@@ -216,8 +216,9 @@ Emit one `security-scan-report.json` — the machine-checkable contract CI gates
 
 In a project with a `02-DOCS/` layer (the [`harness`](../harness/SKILL.md)
 Karpathy wiki), record the scanner choices, pinned versions, gate thresholds, and
-any accepted-risk suppressions in `02-DOCS/wiki/stack/security-scan.md`, and link
-it from the root `CLAUDE.md` `## Knowledge map`. Read it first on every run so the
+any accepted-risk suppressions in `02-DOCS/wiki/stack/security-scan.md`, and index
+it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short
+pointer to it). Read it first on every run so the
 next agent inherits the pinned tools and thresholds instead of re-deriving them.
 No `02-DOCS/`? Skip silently. Conventions are recorded, not gated — never block
 the scan on this.

package/skills/ship/SKILL.md

@@ -228,7 +228,7 @@ When no profile exists, default to L2 and proceed — don't stall the ship to as
 Ship is mostly git actions, but the outcome is recorded so the knowledge model stays whole:
 
 - **The landing decision** (which of the three options, and why) → append to `02-DOCS/wiki/sdd/decisions.md`, the same append-only log `implement`, `verify`, and `review` write to. Parks and discards are logged with their reason so dead-ends aren't re-walked.
-- A **shipped feature** flips its spec under `02-DOCS/wiki/sdd/specs/<slug>.md` to a shipped state (note the merge commit / PR). The harness owns the wiki; ship just keeps the `sdd/` rows in the root `CLAUDE.md` `## Knowledge map` honest.
+- A **shipped feature** flips its spec under `02-DOCS/wiki/sdd/specs/<slug>.md` to a shipped state (note the merge commit / PR). The harness owns the wiki; ship just keeps the `sdd/` rows in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer) honest.
 - An **archive bundle** closes the loop under `02-DOCS/wiki/sdd/archive/<slug>/`:
   - `final-report.md` — what shipped, why, landing decision, links.
   - `apply-progress.md` — copy or link to `progress/<slug>.md`.

package/skills/specify/SKILL.md

@@ -11,7 +11,7 @@ origin: risco
 
 This is the **specify** phase of the rsc-sdd chain: `constitution` → **`specify`** → `clarify` → `plan` → `tasks` → `analyze` → `implement` → `verify` → `review` → `ship`. Its single job is to turn a fuzzy intent into a written specification that states **WHAT** the change is and **WHY** it matters — and nothing about **HOW** it gets built.
 
-A spec is a contract about behaviour and outcomes, readable by a non-technical stakeholder and precise enough that a `plan` can be derived from it. The output is one file: `02-DOCS/wiki/sdd/specs/<slug>.md`, indexed in the root `CLAUDE.md` Knowledge map.
+A spec is a contract about behaviour and outcomes, readable by a non-technical stakeholder and precise enough that a `plan` can be derived from it. The output is one file: `02-DOCS/wiki/sdd/specs/<slug>.md`, indexed in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer).
 
 ## Detect the moment — and hold the gate
 
@@ -93,7 +93,8 @@ Run these in order. It is a collaborative dialogue, not a form you fill in silen
 5. PROPOSE 2-3 approaches → distinct directions with honest trade-offs; lead with your recommendation and why
 6. PRESENT the design     → section by section (problem, users, behaviour, acceptance), scaled to complexity;
                             after EACH section ask "does this look right?" and adjust before moving on
-7. WRITE the spec         → 02-DOCS/wiki/sdd/specs/<slug>.md (WHAT/WHY), index it in CLAUDE.md, commit if a repo
+7. WRITE the spec         → 02-DOCS/wiki/sdd/specs/<slug>.md (WHAT/WHY), index it in 02-DOCS/wiki/index.md
+                            (the Knowledge map; root CLAUDE.md keeps only a short pointer), commit if a repo
 8. SELF-REVIEW            → scan for TODO/placeholder, contradictions, ambiguity, scope creep; fix inline
 9. USER APPROVES          → ask them to read the written spec and confirm; loop on changes until they approve
 10. HAND OFF              → only now, result envelope → clarify/plan. NEVER to implement.
@@ -191,7 +192,7 @@ The proposal is allowed to mention options and tradeoffs; the spec that follows
 - Read `02-DOCS/wiki/sdd/constitution.md` first — its principles are inherited constraints, not things to re-decide. If it's missing, note that the project has no constitution yet and suggest the `constitution` phase before continuing (you can still draft a spec, but flag the absence).
 - **No constitution yet?** Still write the spec, but inherit nothing — lean harder on the wiki and the user's answers, and record every constraint you would have inherited as a *point to clarify* instead of assuming it.
 - Write the spec to `02-DOCS/wiki/sdd/specs/<slug>.md`. Create the directory if absent.
-- Add a row under the `## Knowledge map` section of the root `CLAUDE.md` linking the new spec under the `sdd/specs` topic (additive only — never delete existing rows). Create `CLAUDE.md` if absent.
+- Add a row in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer) linking the new spec under the `sdd/specs` topic (additive only — never delete existing rows). Create the index if absent.
 - Log the spec's creation and any significant scoping decision to `02-DOCS/wiki/sdd/decisions.md` (append-only), so the chain keeps a trace of why scope landed where it did. This is the canonical SDD decisions log shared with `constitution` and `plan` — not the harness's own `02-DOCS/wiki/harness/decisions.md`.
 
 ## Result envelope

package/skills/tasks/SKILL.md

@@ -153,7 +153,8 @@ The task list is **not** a new document. Append it to the existing plan under a
        T001 … T0NN as the table below, + a one-line "generated by tasks on <date>"
 ```
 
-Then ensure the plan is indexed in the root `CLAUDE.md` `## Knowledge map` under
+Then ensure the plan is indexed in `02-DOCS/wiki/index.md` (the Knowledge map;
+root `CLAUDE.md` keeps only a short pointer) under
 the `sdd/` topic (the `plan` phase usually added the row; confirm it points at
 `02-DOCS/wiki/sdd/plans/<slug>.md`, add it if missing — additive only, never
 delete a user's map entry).
@@ -219,7 +220,7 @@ Before handing off, confirm:
 - [ ] Every task traces to a spec line; no orphan tasks.
 - [ ] Dependencies form a valid order (nothing precedes what it needs).
 - [ ] `[P]` markers only on file/state-disjoint tasks.
-- [ ] The list is appended under `## Tasks` in the plan artifact, indexed in the Knowledge map.
+- [ ] The list is appended under `## Tasks` in the plan artifact, indexed in the Knowledge map at `02-DOCS/wiki/index.md` (root `CLAUDE.md` keeps only a short pointer).
 - [ ] A final closer task gates on all done-checks + `verify.sh`.
 - [ ] Review workload forecast and suggested delivery strategy appended.
 

package/skills/verify/SKILL.md

@@ -113,7 +113,7 @@ A criterion you "reviewed by reading the code" is not verified. Reading is `revi
 
 ### 4 — RECORD
 
-Write a dated verification record to `02-DOCS/wiki/sdd/verifications/<slug>-YYYY-MM-DD.md` so the project's living knowledge carries the proof, then ensure it's indexed in the root `CLAUDE.md` `## Knowledge map` under the `sdd/` topic. Keep it short and factual:
+Write a dated verification record to `02-DOCS/wiki/sdd/verifications/<slug>-YYYY-MM-DD.md` so the project's living knowledge carries the proof, then index it in `02-DOCS/wiki/index.md` (the Knowledge map; root `CLAUDE.md` keeps only a short pointer) under the `sdd/` topic. Keep it short and factual:
 
 ```markdown
 # Verification — <slug> — YYYY-MM-DD

package/targets/agents.js

@@ -0,0 +1,98 @@
+// rsc developer agent — installed with the harness for every target that supports
+// file-based subagents with a per-agent model. The agent runs at the `balanced` tier
+// (never `light`/Haiku): Sonnet for Anthropic-backed tools, the provider's mid model
+// elsewhere. The chosen tier (balanced default, or heavy) lives in `.rsc/developer.json`,
+// written by `init` at onboarding and read here so re-syncs honor it.
+import { readFileSync, writeFileSync, mkdirSync, rmSync, existsSync } from 'node:fs';
+import { join, dirname } from 'node:path';
+
+// Concrete model per provider per tier. June 2026 defaults — EDIT to your account's
+// models; the TIER is the contract, the id is yours to change. `light` is deliberately
+// absent: the developer floor is `balanced`.
+const TIER_MODEL = {
+  anthropic: { balanced: 'claude-sonnet-4-6', heavy: 'claude-opus-4-8' },
+  google: { balanced: 'gemini-2.5-flash', heavy: 'gemini-2.5-pro' },
+  openai: { balanced: 'gpt-5.1-mini', heavy: 'gpt-5.1' },
+};
+
+// Per-target agent capability: where the file goes, its format, and how the model value
+// is written for that tool. Targets absent here have no installable file-based agents
+// (Amp/Zed/Windsurf/Cline/Roo/Continue/Aider/Jules/Antigravity) — there the developer
+// model is advisory (model-routing), not a file.
+const AGENT_TARGETS = {
+  claude: { dir: '.claude/agents', ext: '.md', format: 'md', model: (t) => (t === 'heavy' ? 'opus' : 'sonnet') },
+  junie: { dir: '.junie/agents', ext: '.md', format: 'md', model: (t) => (t === 'heavy' ? 'opus' : 'sonnet') },
+  cursor: { dir: '.cursor/agents', ext: '.md', format: 'md', model: (t) => TIER_MODEL.anthropic[t] },
+  opencode: { dir: '.opencode/agents', ext: '.md', format: 'md', mode: 'subagent', model: (t) => `anthropic/${TIER_MODEL.anthropic[t]}` },
+  gemini: { dir: '.gemini/agents', ext: '.md', format: 'md', model: (t) => TIER_MODEL.google[t] },
+  copilot: { dir: '.github/agents', ext: '.agent.md', format: 'md', model: (t) => TIER_MODEL.anthropic[t] },
+  kiro: { dir: '.kiro/agents', ext: '.json', format: 'json', model: (t) => (t === 'heavy' ? 'claude-opus-4' : 'claude-sonnet-4') },
+  codex: { dir: '.codex/agents', ext: '.toml', format: 'toml', model: (t) => TIER_MODEL.openai[t] },
+};
+
+export const AGENT_TARGET_IDS = Object.keys(AGENT_TARGETS);
+export function targetHasAgents(target) { return Boolean(AGENT_TARGETS[target]); }
+
+const NAME = 'developer';
+const DESC = 'Implementation worker: turns an approved spec+plan into working, tested code under strict TDD (red->green->refactor), one task at a time. The rsc SDD fan-out/implementation hand.';
+const BODY = `You are the **developer** subagent for this project — the hands of the rsc SDD chain. You execute a planned, approved task into working, tested code. You do NOT design features.
+
+- Work **test-first**: smallest failing test (RED), least code to pass it (GREEN), then refactor on green. A test that never failed proves nothing.
+- One task at a time; keep the diff to that task's scope — no "while I'm here".
+- Follow the project's spec, plan and constitution under \`02-DOCS/wiki/sdd/\`, and borrow test mechanics from the stack skill (fastapi/go/nextjs/flutter/...).
+- If there is no approved spec + plan for non-trivial feature work, STOP and route to \`specify\` — do not write feature code.
+- Log non-obvious decisions to \`02-DOCS/wiki/sdd/decisions.md\`. Report your diff + test output at the end.
+
+Full discipline lives in the \`implement\` skill.`;
+
+// `.rsc/developer.json` — the chosen tier (balanced default; never light). `init` writes
+// it on the onboarding answer; the installer reads it so every (re)install/sync matches.
+const tierFile = (cwd) => join(cwd, '.rsc', 'developer.json');
+export function readDeveloperTier(cwd) {
+  try {
+    return JSON.parse(readFileSync(tierFile(cwd), 'utf8')).tier === 'heavy' ? 'heavy' : 'balanced';
+  } catch { return 'balanced'; }
+}
+export function writeDeveloperTier(cwd, tier) {
+  const t = tier === 'heavy' ? 'heavy' : 'balanced';
+  mkdirSync(dirname(tierFile(cwd)), { recursive: true });
+  writeFileSync(tierFile(cwd), `${JSON.stringify({ tier: t }, null, 2)}\n`);
+  return t;
+}
+
+function renderMd(spec, model) {
+  const fm = ['---', `name: ${NAME}`, `description: "${DESC}"`, `model: ${model}`];
+  if (spec.mode) fm.push(`mode: ${spec.mode}`);
+  fm.push('---', '');
+  return `${fm.join('\n')}${BODY}\n`;
+}
+const renderJson = (model) => `${JSON.stringify({ name: NAME, description: DESC, model, prompt: BODY }, null, 2)}\n`;
+function renderToml(model) {
+  const esc = (s) => s.replace(/\\/g, '\\\\').replace(/"/g, '\\"');
+  // body as a TOML multiline LITERAL string ('''…''') — no escape processing.
+  return `name = "${NAME}"\ndescription = "${esc(DESC)}"\nmodel = "${model}"\ndeveloper_instructions = '''\n${BODY}\n'''\n`;
+}
+
+export function developerAgentPath(target, cwd) {
+  const spec = AGENT_TARGETS[target];
+  return spec ? join(cwd, ...spec.dir.split('/'), `${NAME}${spec.ext}`) : null;
+}
+
+export function writeDeveloperAgent(target, cwd, tier = readDeveloperTier(cwd)) {
+  const spec = AGENT_TARGETS[target];
+  if (!spec) return [];
+  const model = spec.model(tier);
+  const content = spec.format === 'json' ? renderJson(model)
+    : spec.format === 'toml' ? renderToml(model)
+      : renderMd(spec, model);
+  const path = developerAgentPath(target, cwd);
+  mkdirSync(dirname(path), { recursive: true });
+  writeFileSync(path, content);
+  return [path];
+}
+
+export function removeDeveloperAgent(target, cwd) {
+  const path = developerAgentPath(target, cwd);
+  if (path && existsSync(path)) { rmSync(path, { force: true }); return [path]; }
+  return [];
+}

package/targets/session-start.mjs

@@ -107,6 +107,28 @@ ACTION: run \`npx @ericrisco/rsc audit\`. Opt out with .rsc/.no-audit.
 `);
 }
 
+// CLAUDE.md hygiene: the root CLAUDE.md is read on EVERY turn, so an overgrown one (usually
+// the Knowledge map accreting a row per wiki article) is a permanent context tax that rots
+// adherence. When it passes the ~200-line 2026 budget, nudge to offload the index into
+// 02-DOCS/wiki/index.md and keep CLAUDE.md a short pointer (the `harness` skill owns the move;
+// it's a relocation, not a delete). Opt out with .rsc/.no-claudemd-check.
+const CLAUDEMD_MAX_LINES = 200;
+if (!has('.rsc', '.no-claudemd-check')) {
+  try {
+    const lines = readFileSync(join(root, 'CLAUDE.md'), 'utf8').split('\n').length;
+    if (lines > CLAUDEMD_MAX_LINES) {
+      process.stdout.write(`
+===== rsc CLAUDE.md hygiene =====
+CLAUDE.md is ${lines} lines — over the ~${CLAUDEMD_MAX_LINES}-line budget (it's read every turn, so each line costs context).
+ACTION: offload the Knowledge map / overgrown sections into 02-DOCS/wiki/index.md and leave a short
+pointer in CLAUDE.md (no info lost — it's a move). The \`harness\` skill owns the procedure.
+Opt out with .rsc/.no-claudemd-check.
+=================================
+`);
+    }
+  } catch { /* no CLAUDE.md → nothing to check */ }
+}
+
 // Update check: compare the installed version (.rsc/.version, written at install)
 // against the latest published on npm, and nudge the agent to offer an update.
 // Fail-silent (offline / missing baseline / parse error → nothing). Disable with