@pieerry/harness-kit 3.3.0 → 4.0.0

package/.claude/{plugins → agents}/product-manager/README.md

@@ -1,6 +1,8 @@
-# Product Manager Plugin
+# Product Manager Agent
 
-Native Claude Code plugin for the product team. Drafts PRDs and PRPs with sensor + eval gates, retry loop, token accounting per phase, and optional Confluence publish via hooks.
+Drafts PRDs and PRPs with sensor + eval gates, retry loop, token accounting per phase, and optional Confluence publish via hooks.
+
+Registered in [`AGENTS.md`](../../../AGENTS.md). Agent definition: [`product-manager.md`](../product-manager.md).
 
 ## Slash commands
 
@@ -13,30 +15,30 @@ Also invokable as sub-agent via Task tool with `subagent_type: "product-manager"
 ## Tree
 
 ```
-.claude/plugins/product-manager/
-├── .claude-plugin/plugin.json
-├── agents/product-manager.md sub-agent
-├── commands/run.md orchestrator (calls the two skills)
+.claude/agents/product-manager/      ← definitions
+├── README.md                          this file
 ├── skills/
-│ ├── prd/SKILL.md workflow PRD
-│ └── prp/SKILL.md workflow PRP
-├── hooks/ 5 .sh, registered in .claude/settings.json
-├── scripts/ 4 .py, called by hooks
+│   ├── prd/SKILL.md                   workflow PRD
+│   └── prp/SKILL.md                   workflow PRP
 ├── guides/
-│ ├── pipeline.md retry, approval marker, publish, token accounting
-│ ├── product-guidelines.md team product rules
-│ ├── prd-guidelines.md PRD-specific rules
-│ ├── prp-guidelines.md PRP-specific rules
-│ ├── writing-style.md voice, banned words
-│ ├── templates/{prd,prp}.md artifact templates
-│ └── examples/ reference PRDs and PRPs
-├── sensors/ deterministic checks (markdown)
-├── evals/ LLM-judge rubrics (markdown)
-└── outputs/
- ├── prd/ generated PRDs
- ├── prp/ generated PRPs
- ├── tokens/ per-feature phase tokens JSON
- └── .markers/ phase start/end markers (transient)
+│   ├── pipeline.md                    retry, approval marker, publish, token accounting
+│   ├── product-guidelines.md          team product rules
+│   ├── prd-guidelines.md              PRD-specific rules
+│   ├── prp-guidelines.md              PRP-specific rules
+│   ├── writing-style.md               voice, banned words
+│   ├── templates/{prd,prp}.md         artifact templates
+│   └── examples/                      reference PRDs and PRPs
+├── sensors/                           deterministic checks (markdown)
+└── evals/                             LLM-judge rubrics (markdown)
+
+.claude/runtime/                     ← state + outputs
+├── hooks/product-manager/             5 .sh, registered in .claude/settings.json
+├── scripts/product-manager/           4 .py, called by hooks
+└── outputs/pm/
+    ├── prd/                           generated PRDs
+    ├── prp/                           generated PRPs
+    ├── tokens/                        per-feature phase tokens JSON
+    └── .markers/                      phase start/end markers (transient)
 ```
 
 ## Where to edit
@@ -63,30 +65,30 @@ Also invokable as sub-agent via Task tool with `subagent_type: "product-manager"
 4. PostToolUse hook fires sensor-runner.py with real regex. Blocks on failure, returns feedback. On pass, writes `prd-generate.end` and `prd-validate.start` markers.
 5. Claude applies eval rubric. Retry up to 3 times if score below threshold (rules in guides/pipeline.md).
 6. On pass, Claude appends approval marker. Hook fires: writes `prd-validate.end`, calls token-phase.py for both phases, then confluence-publish.py if creds set.
-7. Token data lands in outputs/tokens/{feature_id}.json. Inline summary comment is appended to the artifact.
+7. Token data lands in .claude/runtime/outputs/pm/tokens/{feature_id}.json. Inline summary comment is appended to the artifact.
 
 ## Token accounting
 
 Per phase: prd-generate, prd-validate, prp-generate, prp-validate.
 
-Per feature, a single file `outputs/tokens/{feature_id}.json` collects all phase entries, with `totals` aggregated. Future workflows (dev, code review) can append their own phases to the same file by reusing feature_id.
+Per feature, a single file `.claude/runtime/outputs/pm/tokens/{feature_id}.json` collects all phase entries, with `totals` aggregated. Future workflows (dev, code review) can append their own phases to the same file by reusing feature_id.
 
 Query examples:
 
 ```
 # total tokens across all features
-jq -s 'map(.totals.input + .totals.output) | add' outputs/tokens/*.json
+jq -s 'map(.totals.input + .totals.output) | add' .claude/runtime/outputs/pm/tokens/*.json
 
 # tokens for one feature, by phase
-jq '.phases[] | {phase, tokens}' outputs/tokens/2026-05-12-billing-tz-fix.json
+jq '.phases[] | {phase, tokens}' .claude/runtime/outputs/pm/tokens/2026-05-12-billing-tz-fix.json
 
 # features touched by dispatch squad
-jq -s '.[] | select(.feature_id | contains("dispatch"))' outputs/tokens/*.json
+jq -s '.[] | select(.feature_id | contains("dispatch"))' .claude/runtime/outputs/pm/tokens/*.json
 ```
 
 ## Engineering handoff
 
-After a PRP is approved, engineering picks it up via the [staff-software-engineer plugin](../staff-software-engineer/README.md). The SSE plugin reads `outputs/prp/{feature_id}.md` and runs plan → dev → test → pr stages, all writing to the same `outputs/tokens/{feature_id}.json` file. Full feature lifecycle in one token log.
+After a PRP is approved, engineering picks it up via the [staff-software-engineer plugin](../staff-software-engineer/README.md). The SSE plugin reads `.claude/runtime/outputs/pm/prp/{feature_id}.md` and runs plan → dev → test → pr stages, all writing to the same `.claude/runtime/outputs/pm/tokens/{feature_id}.json` file. Full feature lifecycle in one token log.
 
 ## Status bar
 

package/.claude/{plugins → agents}/product-manager/guides/examples/good-prp-example.md

@@ -1,6 +1,6 @@
 # PRP: Invoice Deadline Timezone Fix
 
-**Source PRD:** `outputs/prd/2026-05-08-billing-invoice-deadline-tz-fix.md`
+**Source PRD:** `.claude/runtime/outputs/pm/prd/2026-05-08-billing-invoice-deadline-tz-fix.md`
 **Target executor:** coding-agent
 **Squad:** Billing | **Tech lead:** @tech-lead | **Date:** 2026-05-10
 
@@ -122,7 +122,7 @@ psql $STAGING_DB_URL -f src/main/resources/scripts/backfill/invoice_deadline_tz_
 
 ## 9) References
 
-- PRD: `outputs/prd/2026-05-08-billing-invoice-deadline-tz-fix.md`
+- PRD: `.claude/runtime/outputs/pm/prd/2026-05-08-billing-invoice-deadline-tz-fix.md`
 - ADR feature flags: `docs/adr/0014-feature-flags.md`
 - Deadline dashboard: https://grafana.example.com/d/billing-deadline-health
 - Pattern reference: `billing-service/src/main/java/com/example/billing/dunning/DunningService.java:67`

package/.claude/{plugins → agents}/product-manager/guides/pipeline.md

@@ -40,7 +40,7 @@ Triggers hooks/post-eval-{prd|prp}.sh.
 
 ## Publish
 
-Local: file stays in outputs/{prd|prp}/.
+Local: file stays in .claude/runtime/outputs/pm/{prd|prp}/.
 
 Confluence: fires if JIRA_USERNAME and JIRA_API_TOKEN set. Calls scripts/confluence-publish.py. Missing creds skips silently.
 
@@ -52,21 +52,21 @@ After publish, hook appends:
 ## Token accounting
 
 Each phase brackets measurable token window. Phases:
-- prd-generate: from skill invocation to first save in outputs/prd/
+- prd-generate: from skill invocation to first save in .claude/runtime/outputs/pm/prd/
 - prd-validate: from first save to approval marker (sensors + evals)
-- prp-generate: from skill invocation to first save in outputs/prp/
+- prp-generate: from skill invocation to first save in .claude/runtime/outputs/pm/prp/
 - prp-validate: from first save to approval marker
 
-Markers in outputs/.markers/{feature_id}.{phase}.{start|end}, each `{"timestamp": ISO, "session_id": ""}`. Skill writes .start; hooks write .end.
+Markers in .claude/runtime/outputs/pm/.markers/{feature_id}.{phase}.{start|end}, each `{"timestamp": ISO, "session_id": ""}`. Skill writes .start; hooks write .end.
 
-After eval passes, publish hook runs scripts/token-phase.py for both phases. Script reads Claude transcript JSONL, sums usage tokens (input, output, cache_read, cache_creation) within each window, appends phase entry to outputs/tokens/{feature_id}.json. Markers deleted after consumption.
+After eval passes, publish hook runs scripts/token-phase.py for both phases. Script reads Claude transcript JSONL, sums usage tokens (input, output, cache_read, cache_creation) within each window, appends phase entry to .claude/runtime/outputs/pm/tokens/{feature_id}.json. Markers deleted after consumption.
 
 Post-eval hook also appends inline summary to artifact:
 ```
-<!-- tokens: outputs/tokens/{feature_id}.json in={N} out={N} cache_r={N} -->
+<!-- tokens: .claude/runtime/outputs/pm/tokens/{feature_id}.json in={N} out={N} cache_r={N} -->
 ```
 
-Future phases (dev, code review, launch) append entries to same outputs/tokens/{feature_id}.json by reusing feature_id slug.
+Future phases (dev, code review, launch) append entries to same .claude/runtime/outputs/pm/tokens/{feature_id}.json by reusing feature_id slug.
 
 If transcript not readable, script logs warning and exits 0. Token accounting never blocks publish.
 

package/.claude/{plugins → agents}/product-manager/sensors/prp-links.md

@@ -7,7 +7,7 @@ Implemented in scripts/link-validator.py. Same rules when agent self-checks.
 
 ## Hard checks (block)
 
-Source PRD resolvable. "Source PRD:" line must point to path existing under outputs/prd/, relative to PRP file, or relative to repo root.
+Source PRD resolvable. "Source PRD:" line must point to path existing under .claude/runtime/outputs/pm/prd/, relative to PRP file, or relative to repo root.
 
 No localhost URLs. `http://localhost` or `http://127.0.0.1` not allowed as pinned references.
 

package/.claude/{plugins → agents}/product-manager/skills/prd/SKILL.md

@@ -11,7 +11,7 @@ Ask once if missing: squad, problem in 1-2 sentences, customers, hypothesis, bet
 Compute feature_id = {YYYY-MM-DD}-{squad}-{slug}. Before generating, write the phase start marker:
 
 ```
-outputs/.markers/{feature_id}.prd-generate.start
+.claude/runtime/outputs/pm/.markers/{feature_id}.prd-generate.start
 ```
 
 Content: `{"timestamp": "<ISO-8601 UTC now>", "session_id": ""}`
@@ -24,7 +24,7 @@ Read:
 - guides/pipeline.md
 - guides/examples/good-prd-example.md
 
-Save to outputs/prd/{feature_id}.md.
+Save to .claude/runtime/outputs/pm/prd/{feature_id}.md.
 
 Sensors: sensors/prd-structure.md, sensors/prd-acceptance-criteria.md.
 

package/.claude/{plugins → agents}/product-manager/skills/prp/SKILL.md

@@ -6,14 +6,14 @@ user_invocable: true
 
 Generate a PRP. Follow guides/pipeline.md for retry, approval, and publish.
 
-Source PRD: if user passes a path, use it. Else pick the most recent in outputs/prd/. None found, abort. Tell user to run /product-manager:prd first. hooks/pre-prp-check.sh blocks if the PRD lacks the approved marker.
+Source PRD: if user passes a path, use it. Else pick the most recent in .claude/runtime/outputs/pm/prd/. None found, abort. Tell user to run /product-manager:prd first. hooks/pre-prp-check.sh blocks if the PRD lacks the approved marker.
 
-Compute feature_id from the source PRD filename (basename without .md). Save the PRP to outputs/prp/{feature_id}.md so it matches.
+Compute feature_id from the source PRD filename (basename without .md). Save the PRP to .claude/runtime/outputs/pm/prp/{feature_id}.md so it matches.
 
 Before generating, write the phase start marker:
 
 ```
-outputs/.markers/{feature_id}.prp-generate.start
+.claude/runtime/outputs/pm/.markers/{feature_id}.prp-generate.start
 ```
 
 Content: `{"timestamp": "<ISO-8601 UTC now>", "session_id": ""}`
@@ -28,7 +28,7 @@ Read:
 
 Explore target repos. Ask user for repo paths if not provided. Use Grep and Read to map files. Capture file:line. Never invent paths.
 
-Save to outputs/prp/{feature_id}.md.
+Save to .claude/runtime/outputs/pm/prp/{feature_id}.md.
 
 Sensors: sensors/prp-structure.md, sensors/prp-context-quality.md, sensors/prp-links.md.
 

package/.claude/agents/product-manager.md

@@ -7,14 +7,14 @@ model: opus
 
 Product Manager.
 
-When invoked, run /product-manager:run end to end. Follow .claude/plugins/product-manager/guides/pipeline.md.
+When invoked, run /product-manager:run end to end. Follow .claude/agents/product-manager/guides/pipeline.md.
 
 Ask once if missing: team or squad, problem in 1-2 sentences, customers, hypothesis, bet link.
 
 Operating rules:
 - English by default. Domain terms stay native if team uses them.
 - Never invent. Mark gaps with `NOT FOUND - NEEDS REVIEW: {detail}`.
-- Voice: read .claude/plugins/product-manager/guides/writing-style.md. No em-dashes. Mermaid not ASCII.
+- Voice: read .claude/agents/product-manager/guides/writing-style.md. No em-dashes. Mermaid not ASCII.
 - Specific over generic. Real numbers, names, quotes.
 
 Return format: see .claude/commands/product-manager/run.md. Include paths, scores, attempts, publish status, blockers.

package/.claude/agents/staff-software-engineer/README.md

@@ -0,0 +1,87 @@
+# Staff Software Engineer Agent
+
+Drives plan, dev, test, and pr stages across backend, web, mobile, and devops. Per-project conventions in each repo's `.claude/conventions/` override defaults.
+
+Registered in [`AGENTS.md`](../../../AGENTS.md). Agent definition: [`staff-software-engineer.md`](../staff-software-engineer.md).
+
+## Slash commands
+
+- `/sse:plan`: generate an implementation plan from an approved PRP
+- `/sse:dev`: implement the plan in code, run convention gates
+- `/sse:test`: run the project test suite
+- `/sse:pr`: open the draft PR
+- `/sse:run`: full pipeline, plan to pr
+
+Also invokable as sub-agent via Task tool with `subagent_type: "staff-software-engineer"`.
+
+## Tree
+
+```
+.claude/agents/staff-software-engineer/  ← definitions
+├── README.md                              this file
+├── skills/
+│   ├── backend/SKILL.md                   Java/Spring defaults
+│   ├── web/SKILL.md                       Vue/React defaults
+│   ├── mobile/SKILL.md                    iOS/Android defaults
+│   └── devops/SKILL.md                    CI/IaC defaults
+├── guides/
+│   ├── pipeline.md                        retry, approval, token accounting
+│   ├── coding-style.md                    team code style
+│   ├── commit-style.md                    Conventional Commits with TICKET
+│   └── conventions-override.md            how project overrides work
+├── sensors/                               plan, dev, test, pr structure + conventions
+└── evals/                                 plan/dev/test/pr quality rubrics
+
+.claude/runtime/                          ← state + outputs
+├── hooks/staff-software-engineer/         phase markers, sensor gates
+├── scripts/staff-software-engineer/       symlinks to PM scripts (sensor-runner, token-phase)
+└── outputs/sse/
+    ├── plan/                              generated plans
+    ├── dev/                               dev summaries
+    ├── test/                              test results
+    ├── pr/                                opened PR records
+    ├── tokens/                            per-feature phase tokens JSON
+    └── .markers/                          phase start/end markers (transient)
+```
+
+## How conventions work
+
+The plugin holds team defaults per area. Each project repo can override by adding:
+
+```
+{repo-root}/.claude/conventions/{area}.md
+```
+
+Example for the `recon-service` repo:
+
+```
+recon-service/.claude/conventions/backend.md
+```
+
+Plugin skills read both. Project rules win. See `guides/conventions-override.md`.
+
+## Where to edit
+
+| Change | File |
+|--------|------|
+| Pipeline order | guides/pipeline.md |
+| Retry count | guides/pipeline.md (Max attempts) |
+| Plan template/rules | skills/backend/SKILL.md (etc per area), commands/plan.md |
+| Eval threshold | evals/plan-quality.md (Threshold) |
+| Code style | guides/coding-style.md |
+| Commit format | guides/commit-style.md |
+| Test command detection | commands/test.md |
+| PR template | commands/pr.md, hooks/post-eval-pr.sh |
+| Sensors | sensors/*.md |
+
+## Connects to PM plugin
+
+`/sse:run` reads the latest approved PRP from `.claude/runtime/outputs/pm/prp/`. The feature_id flows through: PM plugin creates `2026-05-12-billing-tz-fix`, SSE plugin reuses the same id and writes its phases to the same `.claude/runtime/outputs/sse/tokens/{feature_id}.json` file (per `guides/pipeline.md`).
+
+Full lifecycle in one JSON: PRD generate, PRD validate, PRP generate, PRP validate, plan generate, plan validate, dev, test, pr.
+
+## Setup
+
+Hooks registered in `.claude/settings.json` under PostToolUse. Token script reused from the PM plugin via symlinks in `scripts/`.
+
+PR opening requires the `gh` CLI authenticated. `pr.md` command details ticket detection and template.

package/.claude/{plugins → agents}/staff-software-engineer/guides/conventions-override.md

@@ -1,6 +1,6 @@
 # Conventions Override
 
-How project-specific conventions override plugin defaults.
+How project-specific conventions override agent defaults.
 
 ## Path
 
@@ -26,7 +26,7 @@ When skill or command generates code:
 2. Check if `cwd/.claude/conventions/{area}.md` exists.
 3. If yes, read it. Rules override or add to defaults.
 4. On conflict, project wins.
-5. No project file, use plugin defaults only.
+5. No project file, use agent defaults only.
 
 ## What goes in project conventions file
 
@@ -58,7 +58,7 @@ Keep short. Document what is DIFFERENT, not what is shared with defaults.
 
 ## Avoiding drift
 
-If convention you wrote for repo applies to others, consider promoting to plugin defaults (skills/{area}/SKILL.md). Per-repo files hold genuine repo-specific decisions.
+If convention you wrote for repo applies to others, consider promoting to agent defaults (skills/{area}/SKILL.md). Per-repo files hold genuine repo-specific decisions.
 
 ## Versioning
 

package/.claude/{plugins → agents}/staff-software-engineer/guides/pipeline.md

@@ -40,15 +40,19 @@ Triggers hooks/post-eval-{stage}.sh.
 
 Phases tracked per feature_id (matches PRD/PRP feature_id from PM plugin):
 - plan-generate, plan-validate
-- dev
-- test
-- pr
+- dev-generate, dev-validate
+- test-generate, test-validate
+- pr-generate, pr-validate
 
-Markers in outputs/.markers/{feature_id}.{phase}.{start|end}, each `{"timestamp": ISO, "session_id": ""}`. Skills write .start; hooks write .end. After each phase ends, publish hook calls scripts/token-phase.py.
+Markers in .claude/runtime/outputs/sse/.markers/{feature_id}.{phase}.{start|end}, each `{"timestamp": ISO, "session_id": ""}`.
+Flow per stage:
+1. Skill writes `{stage}-generate.start` before drafting artifact.
+2. Write tool fires post-write-sse.sh → writes `{stage}-generate.end` + `{stage}-validate.start`.
+3. Skill appends approval marker via Edit → fires post-eval-sse.sh → writes `{stage}-validate.end` + runs scripts/token-phase.py for both phases.
 
-Tokens land in shared file with PM plugin: `outputs/tokens/{feature_id}.json`. Same file collects PM phases (prd-generate, prd-validate, prp-generate, prp-validate) and SSE phases (plan-generate, plan-validate, dev, test, pr). Totals cover full feature lifecycle.
+Tokens land in shared file with PM plugin: `.claude/runtime/outputs/sse/tokens/{feature_id}.json`. Same file collects PM phases (prd-generate, prd-validate, prp-generate, prp-validate) and SSE phases (plan-generate, plan-validate, dev, test, pr). Totals cover full feature lifecycle.
 
-To merge with PM tokens file, SSE token-phase.py writes to same path under this plugin's outputs/tokens/, then small step in commands/run.md syncs (or symlinks) with PM tokens dir. v1: SSE keeps own outputs/tokens/{feature_id}.json. v2: merge.
+To merge with PM tokens file, SSE token-phase.py writes to same path under this plugin's .claude/runtime/outputs/sse/tokens/, then small step in commands/run.md syncs (or symlinks) with PM tokens dir. v1: SSE keeps own .claude/runtime/outputs/sse/tokens/{feature_id}.json. v2: merge.
 
 ## Orchestrator order
 
@@ -66,4 +70,4 @@ To merge with PM tokens file, SSE token-phase.py writes to same path under this
 
 ## Conventions override
 
-Before any code generation, read `guides/conventions-override.md` and check `cwd/.claude/conventions/{area}.md`. Project-specific rules win over plugin defaults.
+Before any code generation, read `guides/conventions-override.md` and check `cwd/.claude/conventions/{area}.md`. Project-specific rules win over agent defaults.

package/.claude/{plugins → agents}/staff-software-engineer/sensors/dev-structure.md

@@ -3,7 +3,7 @@
 Type: deterministic
 Mode: hard gate
 
-Validates dev summary doc at `outputs/dev/{feature_id}.md`.
+Validates dev summary doc at `.claude/runtime/outputs/sse/dev/{feature_id}.md`.
 
 ## Required sections
 
@@ -30,7 +30,7 @@ Validates dev summary doc at `outputs/dev/{feature_id}.md`.
 ## Required metadata
 
 Header must declare:
-- Source plan (path to `outputs/plan/{feature_id}.md`)
+- Source plan (path to `.claude/runtime/outputs/sse/plan/{feature_id}.md`)
 - Branch
 - Area (backend | web | mobile | devops)
 - Date

package/.claude/{plugins → agents}/staff-software-engineer/sensors/pr-structure.md

@@ -3,7 +3,7 @@
 Type: deterministic
 Mode: hard gate
 
-Validates PR record at `outputs/pr/{feature_id}.md`.
+Validates PR record at `.claude/runtime/outputs/sse/pr/{feature_id}.md`.
 
 ## Required sections
 
@@ -29,8 +29,8 @@ Validates PR record at `outputs/pr/{feature_id}.md`.
 ## Required metadata
 
 Header must declare:
-- Source plan (path to `outputs/plan/{feature_id}.md`)
-- Source dev (path to `outputs/dev/{feature_id}.md`)
+- Source plan (path to `.claude/runtime/outputs/sse/plan/{feature_id}.md`)
+- Source dev (path to `.claude/runtime/outputs/sse/dev/{feature_id}.md`)
 - Branch
 - Date
 

package/.claude/{plugins → agents}/staff-software-engineer/sensors/test-structure.md

@@ -3,7 +3,7 @@
 Type: deterministic
 Mode: hard gate
 
-Validates test report at `outputs/test/{feature_id}.md`.
+Validates test report at `.claude/runtime/outputs/sse/test/{feature_id}.md`.
 
 ## Required sections
 
@@ -28,7 +28,7 @@ Validates test report at `outputs/test/{feature_id}.md`.
 ## Required metadata
 
 Header must declare:
-- Source dev (path to `outputs/dev/{feature_id}.md`)
+- Source dev (path to `.claude/runtime/outputs/sse/dev/{feature_id}.md`)
 - Framework (maven | gradle | npm | pytest | other)
 - Date
 

package/.claude/agents/staff-software-engineer.md

@@ -7,19 +7,19 @@ model: opus
 
 Staff Software Engineer.
 
-When invoked, run /sse:run end to end. Follow .claude/plugins/staff-software-engineer/guides/pipeline.md.
+When invoked, run /sse:run end to end. Follow .claude/agents/staff-software-engineer/guides/pipeline.md.
 
 Ask once if missing:
-- source PRP path (or read latest from .claude/plugins/product-manager/outputs/prp/)
+- source PRP path (or read latest from .claude/runtime/outputs/pm/prp/)
 - target repo path
 - area: backend, web, mobile, or devops (auto-detect from repo files if possible)
 
 Operating rules:
 - English by default. Domain terms stay native if team uses them.
-- Read project conventions first. Repo's .claude/conventions/{area}.md (if present) overrides plugin defaults. See .claude/plugins/staff-software-engineer/guides/conventions-override.md.
+- Read project conventions first. Repo's .claude/conventions/{area}.md (if present) overrides agent defaults. See .claude/agents/staff-software-engineer/guides/conventions-override.md.
 - Match repo style. Read 3+ similar files before writing.
 - Never invent class names, file paths, helpers. If unknown, `TBD - verify with tech lead`.
-- Voice: read .claude/plugins/staff-software-engineer/guides/coding-style.md and .claude/plugins/staff-software-engineer/guides/commit-style.md. No em-dashes. Mermaid not ASCII.
+- Voice: read .claude/agents/staff-software-engineer/guides/coding-style.md and .claude/agents/staff-software-engineer/guides/commit-style.md. No em-dashes. Mermaid not ASCII.
 - Tests for every feature/bugfix.
 
 Return format: see .claude/commands/sse/run.md. Include paths, scores, gates, PR url, blockers.

package/.claude/commands/pipeline/reset.md

@@ -6,6 +6,6 @@ Reset pipeline.
 
 1. Confirm user wants to abandon active pipeline. Show `python3 .claude/scripts/pipeline.py render` so they see what they drop.
 2. On confirmation: `python3 .claude/scripts/pipeline.py clear`.
-3. Output files under `.claude/plugins/*/outputs/` not deleted. Only in-memory state cleared.
+3. Output files under `.claude/runtime/outputs/` not deleted. Only in-memory state cleared.
 
 Status bar returns to idle after reset.

package/.claude/commands/product-manager/prd.md

@@ -2,7 +2,7 @@
 description: Generate a Product Requirements Document for an team squad. Business-facing artifact. Sensors and evals gate.
 ---
 
-Generate PRD. Follow .claude/plugins/product-manager/guides/pipeline.md for retry, approval, publish.
+Generate PRD. Follow .claude/agents/product-manager/guides/pipeline.md for retry, approval, publish.
 
 Print header card before drafting and footer card after gates run. Format: .claude/scripts/stage-card.md.
 
@@ -11,24 +11,24 @@ Ask once if missing: squad, problem in 1-2 sentences, customers, hypothesis, bet
 Compute feature_id = {YYYY-MM-DD}-{squad}-{slug}. Before generating, write phase start marker:
 
 ```
-.claude/plugins/product-manager/outputs/.markers/{feature_id}.prd-generate.start
+.claude/runtime/outputs/pm/.markers/{feature_id}.prd-generate.start
 ```
 
 Content: `{"timestamp": "<ISO-8601 UTC now>", "session_id": ""}`
 
 Read:
-- .claude/plugins/product-manager/guides/product-guidelines.md
-- .claude/plugins/product-manager/guides/prd-guidelines.md
-- .claude/plugins/product-manager/guides/writing-style.md
-- .claude/plugins/product-manager/guides/templates/prd.md
-- .claude/plugins/product-manager/guides/pipeline.md
-- .claude/plugins/product-manager/guides/examples/good-prd-example.md
+- .claude/agents/product-manager/guides/product-guidelines.md
+- .claude/agents/product-manager/guides/prd-guidelines.md
+- .claude/agents/product-manager/guides/writing-style.md
+- .claude/agents/product-manager/guides/templates/prd.md
+- .claude/agents/product-manager/guides/pipeline.md
+- .claude/agents/product-manager/guides/examples/good-prd-example.md
 
-Save to .claude/plugins/product-manager/outputs/prd/{feature_id}.md.
+Save to .claude/runtime/outputs/pm/prd/{feature_id}.md.
 
-Sensors: .claude/plugins/product-manager/sensors/prd-structure.md, .claude/plugins/product-manager/sensors/prd-acceptance-criteria.md.
+Sensors: .claude/agents/product-manager/sensors/prd-structure.md, .claude/agents/product-manager/sensors/prd-acceptance-criteria.md.
 
-Evals: .claude/plugins/product-manager/evals/prd-quality.md, .claude/plugins/product-manager/evals/prd-readiness.md.
+Evals: .claude/agents/product-manager/evals/prd-quality.md, .claude/agents/product-manager/evals/prd-readiness.md.
 
 After save, reply with this exact shape (name actual sensors/evals/guides that ran, don't abbreviate):
 

package/.claude/commands/product-manager/prp.md

@@ -2,37 +2,37 @@
 description: Generate a Product Requirements Prompt for engineering handoff. Needs an approved PRD. Sensors, link validation, and eval gates.
 ---
 
-Generate PRP. Follow .claude/plugins/product-manager/guides/pipeline.md for retry, approval, publish.
+Generate PRP. Follow .claude/agents/product-manager/guides/pipeline.md for retry, approval, publish.
 
 Print header card before drafting and footer card after gates run. Format: .claude/scripts/stage-card.md.
 
-Source PRD: user passes path, use it. Else pick most recent in .claude/plugins/product-manager/outputs/prd/. None found, abort. Tell user to run /product-manager:prd first. .claude/plugins/product-manager/hooks/pre-prp-check.sh blocks if PRD lacks approved marker.
+Source PRD: user passes path, use it. Else pick most recent in .claude/runtime/outputs/pm/prd/. None found, abort. Tell user to run /product-manager:prd first. .claude/runtime/hooks/product-manager/pre-prp-check.sh blocks if PRD lacks approved marker.
 
-Compute feature_id from source PRD filename (basename without .md). Save PRP to .claude/plugins/product-manager/outputs/prp/{feature_id}.md so it matches.
+Compute feature_id from source PRD filename (basename without .md). Save PRP to .claude/runtime/outputs/pm/prp/{feature_id}.md so it matches.
 
 Before generating, write phase start marker:
 
 ```
-.claude/plugins/product-manager/outputs/.markers/{feature_id}.prp-generate.start
+.claude/runtime/outputs/pm/.markers/{feature_id}.prp-generate.start
 ```
 
 Content: `{"timestamp": "<ISO-8601 UTC now>", "session_id": ""}`
 
 Read:
 - source PRD
-- .claude/plugins/product-manager/guides/prp-guidelines.md
-- .claude/plugins/product-manager/guides/writing-style.md
-- .claude/plugins/product-manager/guides/templates/prp.md
-- .claude/plugins/product-manager/guides/pipeline.md
-- .claude/plugins/product-manager/guides/examples/good-prp-example.md
+- .claude/agents/product-manager/guides/prp-guidelines.md
+- .claude/agents/product-manager/guides/writing-style.md
+- .claude/agents/product-manager/guides/templates/prp.md
+- .claude/agents/product-manager/guides/pipeline.md
+- .claude/agents/product-manager/guides/examples/good-prp-example.md
 
 Explore target repos. Ask user for repo paths if not provided. Use Grep and Read to map files. Capture file:line. Never invent paths.
 
-Save to .claude/plugins/product-manager/outputs/prp/{feature_id}.md.
+Save to .claude/runtime/outputs/pm/prp/{feature_id}.md.
 
-Sensors: .claude/plugins/product-manager/sensors/prp-structure.md, .claude/plugins/product-manager/sensors/prp-context-quality.md, .claude/plugins/product-manager/sensors/prp-links.md.
+Sensors: .claude/agents/product-manager/sensors/prp-structure.md, .claude/agents/product-manager/sensors/prp-context-quality.md, .claude/agents/product-manager/sensors/prp-links.md.
 
-Evals: .claude/plugins/product-manager/evals/prp-quality.md, .claude/plugins/product-manager/evals/prp-context-readiness.md.
+Evals: .claude/agents/product-manager/evals/prp-quality.md, .claude/agents/product-manager/evals/prp-context-readiness.md.
 
 After save, reply with this exact shape (name actual sensors/evals/guides that ran):
 

package/.claude/commands/product-manager/run.md

@@ -5,29 +5,33 @@ description: Run the full PM pipeline. PRD then PRP, with sensor and eval gates
 Run end to end.
 
 1. Invoke /product-manager:prd. Wait for approval marker.
-2. .claude/plugins/product-manager/hooks/pre-prp-check.sh validates PRD.
+2. .claude/runtime/hooks/product-manager/pre-prp-check.sh validates PRD.
 3. Invoke /product-manager:prp using just-approved PRD.
 4. Return summary.
 
-Follow .claude/plugins/product-manager/guides/pipeline.md. Read .claude/agents/product-manager.md for inputs and rules.
+Follow .claude/agents/product-manager/guides/pipeline.md. Read .claude/agents/product-manager.md for inputs and rules.
 
 Return format. Name every sensor, eval, guide that ran. Generic summaries not acceptable — list specifics so user sees what was checked and loaded.
 
 ```
 Pipeline complete.
 
-PRD: .claude/plugins/product-manager/outputs/prd/{path}
+PRD: .claude/runtime/outputs/pm/prd/{path}
   sensors: prd-structure ok, prd-acceptance-criteria ok ({sub-checks})
   eval:    prd-quality {score}/10, prd-readiness {score}/10 (attempts: N)
   guides:  product-guidelines.md, prd-guidelines.md, writing-style.md, templates/prd.md
   refs:    business-info.md, squads/{squad}/context.md
 
-PRP: .claude/plugins/product-manager/outputs/prp/{path}
+PRP: .claude/runtime/outputs/pm/prp/{path}
   sensors: prp-structure ok, prp-context-quality ok, prp-links ok
   eval:    prp-quality {score}/10, prp-context-readiness {score}/10 (attempts: N)
   guides:  prp-guidelines.md, writing-style.md, templates/prp.md
   refs:    prd/{feature_id}.md, target repo paths probed
 
+Tokens: outputs/tokens/{feature_id}.json
+  totals: in={N} out={N} cache_r={N}
+  phases: prd-generate, prd-validate, prp-generate, prp-validate
+
 Confluence: {published to {space-key}: {url} | skipped, reason}
 
 Blockers: