loki-mode 11.74.2 → 11.75.1

package/skills/sdlc-fleet.md

@@ -1,170 +0,0 @@
-# SDLC Fleet Pattern (skill module)
-
-<!-- TOC:START (generated by scripts/skill-toc.sh; do not edit) -->
-## Contents
-
-- [When to invoke this pattern](#when-to-invoke-this-pattern)
-- [The 6 roles](#the-6-roles)
-  - [1. Architect (1 agent, Plan subagent_type, model: opus)](#1-architect-1-agent-plan-subagent_type-model-opus)
-  - [2. Product Owner (the integrator, NOT a separate agent)](#2-product-owner-the-integrator-not-a-separate-agent)
-  - [3. Dev Fleet (3-5 principal engineers in parallel, model: opus)](#3-dev-fleet-3-5-principal-engineers-in-parallel-model-opus)
-  - [4. SDET (1-2 agents, model: opus)](#4-sdet-1-2-agents-model-opus)
-  - [5. Council Reviewers (3 agents, parallel, model: 2 opus + 1 sonnet)](#5-council-reviewers-3-agents-parallel-model-2-opus--1-sonnet)
-  - [6. Real-User QA (1 agent OR the integrator directly)](#6-real-user-qa-1-agent-or-the-integrator-directly)
-- [Parallel-execution rules](#parallel-execution-rules)
-- [Honest acknowledgements pattern](#honest-acknowledgements-pattern)
-- [Reference run: v7.6.0 LSP integration](#reference-run-v760-lsp-integration)
-- [Per-release checklist (always)](#per-release-checklist-always)
-<!-- TOC:END -->
-
-**Established:** v7.7.4 (2026-05-24)
-**Origin:** user mandate "agent teams with senior most developers, Loki architects, sdet team and council of reviewers exist and coordinate and work in parallel"
-**Proven by:** v7.6.0 fleet ship (architect + 3 principal engineers + SDET + 3-reviewer council + real-user QA) demonstrated the pattern works end-to-end.
-
-This skill describes the STANDING pattern for shipping any non-trivial change in Loki Mode. The integrator (Claude Code session) is responsible for orchestrating the roles below. Each role is a Loki Agent call.
-
-## When to invoke this pattern
-
-Apply this pattern when shipping ANY of:
-
-- A new feature touching > 3 files
-- A bug fix touching agent runtime, council, memory, or auto-spawn surfaces
-- A MINOR or MAJOR release
-- A change with cross-route parity implications (bash + Bun + TS)
-- A change the user has marked "critical" or "production-blocking"
-
-Skip this pattern (acceptable shortcuts) when:
-
-- Single-line typo fix in user-facing strings
-- Pure docs edit (no code change)
-- A revert of a previously-shipped change with no new logic
-- An emergency hotfix where waiting on parallel agents would exceed the production-impact window
-
-## The 6 roles
-
-### 1. Architect (1 agent, Plan subagent_type, model: opus)
-
-**Job:** Design the change end-to-end before any code is written. Produces a written plan saved as `docs/<TOPIC>-PLAN.md`.
-
-**Inputs:** the original user directive verbatim + relevant source files + any prior council findings.
-
-**Outputs:** a markdown plan with: scope, acceptance criteria, implementation steps, file-level diffs sketch, testing plan, risks + mitigations, NOT-tested honest disclosure.
-
-**Output is BINDING** for the dev fleet -- don't change scope without re-architecting.
-
-### 2. Product Owner (the integrator, NOT a separate agent)
-
-**Job:** Lock scope before implementation. Use `AskUserQuestion` to surface ambiguity to the user. NEVER guess at scope.
-
-**Output:** explicit user answer that locks the acceptance criteria.
-
-### 3. Dev Fleet (3-5 principal engineers in parallel, model: opus)
-
-**Job:** Implement independent slices of the architect's plan IN PARALLEL. Each engineer gets a single self-contained task with explicit file references and binding constraints.
-
-**Pattern:** spawn N agents via `Agent` tool with `run_in_background: true` in ONE message (parallel). Each agent gets:
-
-- The architect's plan section relevant to their slice
-- A SINGLE deliverable (one fix, one feature, one file)
-- Binding constraints (no version bumps, no commits, no emojis, no em dashes, NO destructive operations)
-- Explicit "report back in N words" cap
-
-The integrator does NOT delegate synthesis -- they integrate the parallel outputs themselves.
-
-### 4. SDET (1-2 agents, model: opus)
-
-**Job:** Write tests + capture screenshots. For each shipped feature:
-
-- One or more bash / Playwright tests asserting acceptance criteria
-- For UI surfaces: screenshot capture with `chrome --headless --screenshot` or Playwright
-- For backends: curl-based smoke tests
-
-**Output:** test files + screenshot artifacts under `artifacts/<release>-screens/`.
-
-### 5. Council Reviewers (3 agents, parallel, model: 2 opus + 1 sonnet)
-
-**Job:** Independent code review. Binding gate: unanimous APPROVE required to ship.
-
-**Pattern:** spawn 3 agents in parallel via `Agent` tool. Each gets:
-
-- The full file diff at HEAD
-- Their specialized focus (e.g. "frontend correctness", "backend / integration risk", "release readiness")
-- Strict output format: `VOTE: APPROVE | CONCERN | REJECT` + findings list + reasoning
-
-On any CONCERN or REJECT:
-
-1. Integrator reads source + validates concern
-2. If valid: dispatch fix agent + RE-RUN entire council on post-fix state
-3. If invalid: refute with evidence + re-spawn ONLY the dissenting reviewer with the refutation
-4. Loop until 3-of-3 APPROVE
-5. "2-of-3 is good enough" is NEVER acceptable
-
-### 6. Real-User QA (1 agent OR the integrator directly)
-
-**Job:** Install the just-shipped release fresh from npm/bun, run a user-realistic scenario end-to-end, capture screenshots, report any breakage.
-
-**Pattern:** after `gh release` confirms, run:
-
-```bash
-bun install -g loki-mode@<NEW_VERSION>
-loki --version  # confirm new version on PATH
-# Then exercise the new feature in a user-realistic way:
-loki <new-command> ...  # the change we just shipped
-```
-
-Capture:
-- Exit codes
-- Output structure
-- Any unexpected errors / warnings
-- Browser screenshots if UI surface
-
-## Parallel-execution rules
-
-When spawning multiple agents in the SAME message:
-
-- Use `Agent` tool with `run_in_background: true`
-- Send all N tool calls in a SINGLE message (parallel scheduling)
-- Each agent gets a SELF-CONTAINED prompt (no shared context)
-- Each agent gets a SHORT response cap (under 250 words)
-- The integrator INTEGRATES the parallel outputs -- never delegates synthesis
-
-Anti-pattern: spawning agents sequentially when work is independent. Always parallel.
-
-## Honest acknowledgements pattern
-
-Every CHANGELOG entry that ships work via this pattern MUST include:
-
-```markdown
-### NOT tested in this release
-
-- <each test that was skipped + reason>
-- <any feature shipped without end-to-end real-user repro>
-- <any divergence from architect plan + workaround>
-```
-
-Failing to disclose is fabrication. CLAUDE.md MEMORY.md "Be real, no fabrication, no lying" applies.
-
-## Reference run: v7.6.0 LSP integration
-
-Demonstrated this pattern works:
-
-1. **Architect** (Plan agent): designed Merge-3 in `docs/MERGE3-PLAN.md`
-2. **Product Owner** (integrator + AskUserQuestion): user chose merge scope
-3. **Dev Fleet** (3 principal engineers in parallel): Engineer A fixed memory PYTHONPATH; Engineer B added USAGE.md auto-gen across bash + TS; Engineer C added Memory drill-down UI
-4. **SDET** (Explore + Playwright): captured 20+ UI screenshots, ran 24-scenario CLI permutation test
-5. **Council** (2 Opus + 1 Sonnet): 2 rounds; round 1 had 2 CONCERN-major; fixes applied; round 2 unanimous APPROVE
-6. **Real-user QA**: fresh `bun install -g loki-mode@7.7.0` + ran on a different stack (Python Flask) than dev test + verified USAGE.md auto-generated + Memory drill-down rendered in a real browser
-
-Result: 4 releases shipped that day with all 8 GitHub workflows SUCCESS on each.
-
-## Per-release checklist (always)
-
-The binding release procedure is the **Release Workflow** section of `CLAUDE.md`,
-which is the authoritative and maintained copy: version bump across every listed
-file, dashboard build, tests, pre-publish validation, commit and push, and
-distribution verification across all channels.
-
-This previously pointed at an artifacts/ROADMAP file which has never
-existed in this repository -- so an agent following it for the "binding" steps
-found nothing and had to guess. The steps live in one place now, and that place
-is kept current because every release edits it.

package/tools/cockpit/SPEC.md

@@ -1,275 +0,0 @@
-# CEO cockpit: specification
-
-Status: spec only (owner: platform-lead to build; chief-of-staff to accept).
-Source of the ask: docs/v10/RETRO-2026.md section 6 item 6 (CEO console) and
-item 7 (ROI ledger).
-
-New-file justification (CLAUDE.md "extend, never duplicate"): no existing file
-owns company operations for the founder. `dashboard/server.py` is the
-user-facing per-run product dashboard and must stay free of company state;
-`scripts/v10-pulse.sh`, `tools/cost-history.py` and `scripts/usage-governor.py`
-already own pulse, run cost and usage, so the cockpit only invokes them and
-renders their output. It re-implements none of their logic.
-
-## 1. Goals and non-goals
-Goals:
-- One page where the CEO sees, within 5 seconds of opening, what the company
-  is doing, what shipped, what is blocked and what it costs against value.
-- One write path: a founder command box that appends a row to
-  docs/v10/FOUNDER-QUEUE.md for the chief-of-staff to act on.
-Non-goals:
-- No direct control. Stop-all, dispatch and release actions are queue rows
-  ("STOP ALL", "pause release") that the chief-of-staff executes by recorded
-  PID or through release-reliability-lead. The cockpit never signals a
-  process, edits BOARD.md, runs git writes or calls npm publish.
-- No hosted mode, no auth provider, no telemetry.
-
-## 2. Hard constraints
-1. Server: Python 3 standard library only (`http.server`,
-   `ThreadingHTTPServer`, `json`, `subprocess`, `fcntl`, `secrets`). No pip
-   dependency.
-2. Client: one static file `tools/cockpit/index.html` with inline CSS and JS.
-   No build step, no CDN, no framework. Works offline.
-3. Bind 127.0.0.1 only. The `--host` flag accepts only `127.0.0.1` or `::1`;
-   any other value exits 64 with a message. Never `0.0.0.0`.
-4. Read-only except `POST /api/queue`, which only appends to FOUNDER-QUEUE.md.
-5. Testable headless: `--once --json` prints the full snapshot and exits; no
-   browser is ever opened (honour LOKI_NO_BROWSER; no `webbrowser` import).
-6. Every shell-out has a timeout (default 20 s; pulse 60 s) and a cache, so a
-   hung `gh` or `npm` never blocks the page. A failed source renders
-   UNKNOWN with the error line, never a blank or a 0.
-7. No emojis, no em or en dashes in code, UI text or output.
-
-## 3. Files
-- `tools/cockpit/server.py`: server, collectors, snapshot builder, CLI.
-- `tools/cockpit/index.html`: the page.
-- `tools/cockpit/roi-basis.json`: the salary basis and ROI assumptions (section
-  5.6). Tracked, edited by hand or by the chief-of-staff, versioned in git.
-- `tests/test-cockpit.sh`: headless test, registered in
-  `tests/run-all-tests.sh` and the shard map (retro 3.3.21).
-- Runtime only (untracked, under `.loki/`): `.loki/company/status/<slug>.json`
-  (written by each seat), `.loki/cockpit/cache/*.json`, `.loki/cockpit/token`.
-
-## 4. CLI
-```
-python3 tools/cockpit/server.py [--port 8787] [--host 127.0.0.1]
-                                [--repo-root PATH] [--once] [--json]
-                                [--panel NAME]
-```
-- `--once --json`: build one snapshot, print it, exit 0 (exit 2 if any panel
-  is UNKNOWN, mirroring the pulse contract 0 clean, 1 violation, 2 unknown;
-  exit 1 if the pulse reports a VIOLATION).
-- `--panel NAME`: restrict `--once` to one panel (pulse, releases, board,
-  agents, okrs, roi, queue).
-- Env overrides (for tests and fixtures): `COCKPIT_NPM_CMD`, `COCKPIT_GH_CMD`,
-  `COCKPIT_PULSE_CMD`, `BOARD_MD`, `FOUNDER_QUEUE_MD`, `RETRO_MD`,
-  `USAGE_READINGS_TSV`, `COST_HISTORY_FILE`, `COCKPIT_STATUS_DIR`,
-  `COCKPIT_NOW` (ISO8601 UTC, for deterministic ages).
-- On start the server prints `cockpit: http://127.0.0.1:<port>/` and its PID,
-  and writes the PID to `.loki/cockpit/server.pid` so it is stopped by
-  recorded PID only.
-
-## 5. Panels and data sources
-Each panel in the snapshot has the shape
-`{status: OK|VIOLATION|UNKNOWN|UNMEASURED, as_of, source_cmd, data, error}`.
-`source_cmd` is the exact command or file read, shown on the page under each
-panel (evidence rule, retro 3.1.1).
-
-### 5.1 Pulse
-- Source: `bash scripts/v10-pulse.sh` (override `COCKPIT_PULSE_CMD`), run with
-  `PULSE_REPO_ROOT` set to the repo root. Cached 90 s; refreshed in a
-  background thread so a request never waits on it.
-- Parse: every `Key: value` line into `metrics`; every line under
-  `=== VIOLATIONS ===` starting `VIOLATION:` into `violations`; the
-  `UNKNOWN metrics:` and `STALE metrics` lines into lists. Exit code 0, 1 or 2
-  maps to OK, VIOLATION, UNKNOWN.
-- Render: violations first in a red band, top one largest; then the metric
-  table; STALE values shown with their age.
-
-### 5.2 Releases
-- npm: `npm view loki-mode time dist-tags --json` (override
-  `COCKPIT_NPM_CMD`). Show `dist-tags.latest`, `dist-tags.next`, the last 20
-  versions with publish time, releases in the last hour and 24 h, and minutes
-  since the last publish.
-- Release runs: `gh run list --workflow release.yml --limit 20 --json
-  databaseId,status,conclusion,headSha,createdAt,displayTitle,url` (override
-  `COCKPIT_GH_CMD`). Show each run with conclusion and link.
-- Local: `VERSION` file and `git -C <root> describe --tags --abbrev=0`.
-- Derived flags: VERSION vs next vs latest divergence (retro 3.2.27); a
-  release run with conclusion failure on the newest VERSION SHA is a
-  VIOLATION. A cached packument is not proof: the panel labels npm data
-  "registry view" and never claims a tarball is fetchable (retro 3.2.7).
-
-### 5.3 BOARD flow
-- Source: `docs/v10/BOARD.md` (override `BOARD_MD`). Parse table rows with
-  columns `ID | Owner | Branch @ SHA | File set | Tier | Status | Notes`.
-  Status cell is `<token>@YYYY-MM-DDTHH:MMZ`; tokens: ready, building, review,
-  review-blocked, approved, merged, released, blocked, rejected, parked.
-- Render: count per token (flow strip ready, building, review, approved,
-  merged, released), the active rows (building, review, review-blocked,
-  approved, blocked) with age in minutes against their tier budget (LOW 15,
-  MEDIUM 30, HIGH 60 per OPERATING-MODEL.md), and an "unparsed rows" list.
-  Ready count below 30 is flagged (D91).
-
-### 5.4 Agents
-- Source: `.loki/company/status/<slug>.json` for the 10 slugs in
-  `.claude/agents/` (list derived from that directory, never hard-coded).
-  Schema: `{seat, model, doing, slice, since, next, blocked_on, updated}`.
-- Render one card per seat: model, what it is doing, slice ID linking to the
-  BOARD row, minutes since `updated`. A seat with no file is UNKNOWN; a file
-  older than 60 min is STALE; `blocked_on` non-empty is highlighted.
-- Also show the seat's owned root causes, parsed from the "Root causes
-  owned" section of its agent file.
-
-### 5.5 OKRs
-- Source: `docs/v10/RETRO-2026.md` section 6 (override `RETRO_MD`): the 20
-  numbered items, each parsed into title (first sentence), owner (text after
-  `Owner:` up to the next period) and metric (text after `Metric:`).
-- Current values: `.loki/company/okrs.json`, written by the owning seats,
-  shape `{"<item number>": {value, target, status, source_cmd, as_of}}`.
-- Render the 20 rows ranked as in the retro. An item with no current value
-  shows UNMEASURED, never 0 and never green. A value without `source_cmd` is
-  shown as UNCITED and counts against verification-scientist.
-
-### 5.6 ROI
-Two sides, always labelled with their measurement state.
-
-Agent cost (left side):
-- Plan cost: `roi-basis.json` field `plan_usd_per_month` (default 200, the
-  founder's stated spend, retro RC9 evidence), prorated per day.
-- Usage share: `docs/v10/usage-readings.tsv` (columns `utc_time
-  window_percent weekly_percent`) and `python3 scripts/usage-governor.py
-  --json`. These are plan-usage percentages, not tokens or dollars; the panel
-  says so and shows weekly percent used.
-- Metered run cost: `python3 tools/cost-history.py report --json` (file
-  override `COST_HISTORY_FILE`). Shows measured runs, unmeasured runs (usd
-  null) as a separate count, and total measured USD. A missing history file
-  is UNMEASURED.
-- Budget line: the $50 per day budget from FOUNDER-QUEUE row 3, as a
-  configurable `daily_budget_usd`.
-- Cost is labelled ESTIMATE when derived from plan proration, MEASURED only
-  when it comes from cost-history measured rows.
-
-Human-equivalent value (right side):
-- Salary basis in `roi-basis.json`, versioned, every figure with source URL
-  and snapshot date:
-  ```json
-  {
-    "source_url": "https://www.levels.fyi/companies/anthropic/salaries/software-engineer",
-    "snapshot_date": "2026-09-24",
-    "annual_total_comp_usd": {
-      "entry": 483000, "senior": 621000, "lead": 865000,
-      "staff": 1410000, "median": 883000
-    },
-    "research_scientist_median_usd": {"value": null, "status": "UNVERIFIED"},
-    "research_engineer_usd": {"value": null, "status": "NOT FOUND"},
-    "overhead_multiplier": {"value": 1.25, "status": "ASSUMPTION"},
-    "work_hours_per_year": {"value": 2000, "status": "ASSUMPTION"},
-    "seat_tier": {"chief-of-staff": "lead", "cto": "staff", "...": "..."},
-    "hours_per_outcome": {"verified_slice": {"value": null, "status": "ASSUMPTION, unset"}},
-    "plan_usd_per_month": 200,
-    "daily_budget_usd": 50
-  }
-  ```
-  The figures match RETRO-2026.md 3.8.10 and were re-fetched from the URL on
-  2026-10-23. The basis is flagged STALE when `snapshot_date` is older than
-  90 days.
-- Value formula, shown on the page: hourly rate = annual total comp x
-  overhead multiplier / work hours per year; human-equivalent value = sum over
-  released slices (BOARD released tokens in the window) of hours per outcome
-  x hourly rate for the owning seat's tier.
-- Value is UNMEASURED while `hours_per_outcome` is unset. The page then shows
-  only "standing 10-seat payroll equivalent" (sum of tier comp x overhead for
-  10 seats, labelled ASSUMPTION) next to agent cost, and never prints an ROI
-  multiple.
-- ROI multiple is shown only when both sides are non-null, with the label of
-  the weaker side (ESTIMATE or ASSUMPTION). Per verified outcome cost comes
-  from the outcome scoreboard (`.loki/company/scoreboard/*.json`) when it
-  exists, else UNMEASURED.
-- Every assumption is editable only in `roi-basis.json`; the page shows its
-  path and git SHA, never an inline editor.
-
-### 5.7 Founder command box
-- UI: one textarea (max 2000 characters), one "Why only you" optional field,
-  one submit button. After submit, the page shows the appended row number
-  and the line written.
-- `POST /api/queue`, body JSON `{item, why}`; returns `{row, line}`.
-- Append rules:
-  1. Take an exclusive `fcntl.flock` on `FOUNDER-QUEUE.md.lock` (sidecar,
-     under `.loki/cockpit/`), with a 5 s timeout returning 503.
-  2. Read the file; allocate the next number as max existing `#` + 1.
-  3. Sanitize: strip control characters, replace newlines with spaces and
-     `|` with `/`, trim, reject empty (400) or over the cap (413).
-  4. Write one row in the existing header shape
-     `| # | Date | Item | Why only you | Loop does meanwhile |` with Date as
-     today UTC `YYYY-MM-DD`, Item prefixed `[cockpit]`, Why from the field or
-     `founder command`, Loop cell `chief-of-staff to triage`, plus trailing
-     status cell `open`.
-  5. Append only (`open(path, "a")`), then fsync, then re-read and verify the
-     file did not shrink and the new row is the last line (shrink guard,
-     retro 3.3.17). On failure return 500 and leave the file unchanged.
-- The cockpit never commits the change; the chief-of-staff triages and
-  commits.
-
-## 6. HTTP API
-| Method | Path | Returns |
-|---|---|---|
-| GET | `/` | index.html |
-| GET | `/api/snapshot` | all panels |
-| GET | `/api/panel/<name>` | one panel |
-| GET | `/healthz` | `{"ok": true, "pid": N}` |
-| POST | `/api/queue` | append result |
-Every other method or path returns 405 or 404.
-
-Security:
-- Reject any request whose `Host` header is not `127.0.0.1:<port>`,
-  `localhost:<port>` or `[::1]:<port>` (DNS rebinding).
-- `POST /api/queue` requires header `X-Cockpit-Token` equal to a per-process
-  random token (`secrets.token_urlsafe(32)`), written to
-  `.loki/cockpit/token` with mode 600 and injected into the served page; and
-  an `Origin` equal to the served origin. Missing or wrong: 403.
-- Request body capped at 8 KB. Responses set `Cache-Control: no-store`,
-  `X-Content-Type-Options: nosniff` and a CSP of `default-src 'self';
-  script-src 'self' 'unsafe-inline'; style-src 'self' 'unsafe-inline'`.
-- Subprocesses run with an argument list (never `shell=True`) and an env built
-  deny-by-default (PATH, HOME, USER, LOGNAME, LANG, gh and npm config paths
-  only), per security-lead rule 3.9.2.
-- No file path from a request is ever opened; all paths come from flags or
-  env.
-
-## 7. Page behaviour
-- Polls `/api/snapshot` every 15 s; panels update in place without layout
-  shift (fixed card heights, skeleton rows); a "last refreshed" time and the
-  age of each source are always visible.
-- Order: pulse violations band, agents, BOARD flow, releases, OKRs, ROI,
-  command box.
-- Light and dark from `prefers-color-scheme`; readable at 1280 px and 390 px.
-- Every number has a hover or caption with its `source_cmd` and `as_of`.
-
-## 8. Headless test (`tests/test-cockpit.sh`)
-Uses `loki_run_tmp_create` and `loki_run_tmp_cleanup`, stops the server by its
-recorded PID only, sets `LOKI_NO_BROWSER=1`. Fixtures under the run tmp dir:
-- A BOARD.md with one row per token plus one unparsed row; assert counts.
-- A FOUNDER-QUEUE.md copy; POST an item containing `|`, a newline and 2001
-  characters; assert 413, then a valid item, assert row number = max + 1, the
-  pipe replaced, the file only grew, and two concurrent POSTs get distinct
-  numbers.
-- Stub `COCKPIT_NPM_CMD` and `COCKPIT_GH_CMD` scripts emitting fixed JSON,
-  and one that sleeps past the timeout; assert UNKNOWN, not a hang.
-- Missing cost-history and empty status dir: assert ROI cost UNMEASURED and
-  agents UNKNOWN, and that no panel prints `0` for an unmeasured value.
-- `--host 0.0.0.0` exits 64. POST without token: 403. Bad Host: 403.
-- `--once --json` output parses and its exit code matches the pulse stub's
-  exit code mapping.
-- Positive control for every absence assertion (retro 3.1.6): the same probe
-  run on a fixture that does contain the value must find it.
-
-## 9. Acceptance (Wall checks)
-1. `python3 tools/cockpit/server.py --once --json | python3 -m json.tool`
-   exits 0 or 2 on a clean checkout, with all 7 panels present.
-2. `bash tests/test-cockpit.sh` passes, and fails when the append sanitizer
-   or the Host check is removed (mutation).
-3. `lsof -nP -iTCP:<port> -sTCP:LISTEN` shows 127.0.0.1 only.
-4. No file under the repo changes after an hour of page polling except
-   FOUNDER-QUEUE.md on submit (`git status --porcelain` diff).
-5. Screenshot at 1280 px and 390 px reviewed by developer-experience-lead.

package/tools/cockpit/index.html

@@ -1,146 +0,0 @@
-<!doctype html>
-<html lang="en">
-<head>
-<meta charset="utf-8">
-<meta name="viewport" content="width=device-width, initial-scale=1">
-<title>Loki Cockpit</title>
-<style>
-:root{--bg:#f4f5f8;--card:#fff;--ink:#14161c;--mute:#5b6270;--line:#d9dde6;--ok:#17653a;--bad:#a4161a;--warn:#8a5a00;--unk:#4a4f6a;--badbg:#fde8e8;--accent:#2b4ed8}
-@media (prefers-color-scheme:dark){:root{--bg:#0f1115;--card:#171a21;--ink:#e8eaf0;--mute:#9aa3b5;--line:#2a2f3a;--ok:#5fd08f;--bad:#ff8083;--warn:#f0b955;--unk:#9fa6c8;--badbg:#3a1618;--accent:#8aa2ff}}
-*{box-sizing:border-box}
-body{margin:0;background:var(--bg);color:var(--ink);font:14px/1.45 system-ui,-apple-system,Segoe UI,sans-serif}
-header{display:flex;flex-wrap:wrap;gap:8px 16px;align-items:baseline;padding:12px 16px;border-bottom:1px solid var(--line)}
-header h1{font-size:16px;margin:0}
-#refreshed{color:var(--mute);font-size:12px}
-main{display:grid;grid-template-columns:repeat(auto-fit,minmax(min(100%,420px),1fr));gap:12px;padding:12px 16px}
-section{background:var(--card);border:1px solid var(--line);border-radius:8px;padding:12px;min-height:160px;overflow:hidden}
-section.wide{grid-column:1/-1}
-h2{font-size:13px;margin:0 0 8px;text-transform:uppercase;letter-spacing:.04em;display:flex;gap:8px;align-items:center}
-.badge{font-size:11px;padding:1px 6px;border-radius:10px;border:1px solid currentColor}
-.OK{color:var(--ok)}.VIOLATION{color:var(--bad)}.UNKNOWN,.UNMEASURED,.UNCITED{color:var(--unk)}.STALE{color:var(--warn)}
-.src{margin-top:8px;color:var(--mute);font:11px ui-monospace,Menlo,monospace;word-break:break-all}
-.err{color:var(--warn);font-size:12px;margin:4px 0}
-.band{background:var(--badbg);color:var(--bad);border:1px solid var(--bad);border-radius:6px;padding:8px;margin-bottom:8px}
-.band .top{font-size:16px;font-weight:600}
-table{width:100%;border-collapse:collapse;font-size:12px}
-td,th{text-align:left;padding:3px 6px 3px 0;border-bottom:1px solid var(--line);vertical-align:top;word-break:break-word}
-.flow{display:flex;flex-wrap:wrap;gap:6px;margin-bottom:8px}
-.flow div{border:1px solid var(--line);border-radius:6px;padding:4px 8px;min-width:78px}
-.flow b{display:block;font-size:18px}
-.cards{display:grid;grid-template-columns:repeat(auto-fill,minmax(210px,1fr));gap:8px}
-.card{border:1px solid var(--line);border-radius:6px;padding:8px;font-size:12px}
-.card.blocked{border-color:var(--bad)}
-.mute{color:var(--mute)}
-textarea,input[type=text]{width:100%;background:var(--bg);color:var(--ink);border:1px solid var(--line);border-radius:6px;padding:8px;font:inherit}
-button{background:var(--accent);color:#fff;border:0;border-radius:6px;padding:8px 14px;font:inherit;cursor:pointer}
-button:disabled{opacity:.5}
-#qres{margin-top:8px;font-size:12px;min-height:18px;word-break:break-word}
-</style>
-</head>
-<body>
-<header><h1>Loki Cockpit</h1><span id="refreshed">loading</span><span class="mute" id="fail"></span></header>
-<main>
-<section class="wide" id="p-pulse"><h2>Pulse</h2><div class="body">loading</div></section>
-<section class="wide" id="p-agents"><h2>Agents</h2><div class="body">loading</div></section>
-<section id="p-board"><h2>BOARD flow</h2><div class="body">loading</div></section>
-<section id="p-releases"><h2>Releases</h2><div class="body">loading</div></section>
-<section class="wide" id="p-okrs"><h2>OKRs</h2><div class="body">loading</div></section>
-<section class="wide" id="p-roi"><h2>ROI</h2><div class="body">loading</div></section>
-<section class="wide" id="p-queue"><h2>Founder command box</h2><div class="body">loading</div>
-<div>
-<textarea id="item" rows="3" maxlength="2000" placeholder="Command or decision for the chief of staff (max 2000 characters)"></textarea>
-<input type="text" id="why" maxlength="500" placeholder="Why only you (optional)">
-<p><button id="send">Append to FOUNDER-QUEUE</button> <span class="mute" id="count">0/2000</span></p>
-<div id="qres"></div>
-</div>
-</section>
-</main>
-<script>
-(function(){
-"use strict";
-var TOKEN="__COCKPIT_TOKEN__";
-function h(tag,attrs){var e=document.createElement(tag);if(attrs){for(var k in attrs){if(k==="class")e.className=attrs[k];else if(k==="title")e.title=attrs[k];else e.setAttribute(k,attrs[k]);}}
- for(var i=2;i<arguments.length;i++){var c=arguments[i];if(c===null||c===undefined)continue;e.appendChild(typeof c==="object"?c:document.createTextNode(String(c)));}return e;}
-function show(v){return v===null||v===undefined?"UNKNOWN":String(v);}
-function badge(s){return h("span",{"class":"badge "+s},s);}
-function src(p){return h("div",{"class":"src",title:"as of "+p.as_of},"source: "+p.source_cmd+"  as of "+p.as_of+(p.cache_age_s!==undefined?"  cached "+p.cache_age_s+"s":""));}
-function err(p){return p.error?h("div",{"class":"err"},p.error):null;}
-function table(rows,head){var t=h("table");if(head){var tr=h("tr");head.forEach(function(x){tr.appendChild(h("th",null,x));});t.appendChild(tr);}
- rows.forEach(function(r){var tr=h("tr");r.forEach(function(c){tr.appendChild(h("td",null,c));});t.appendChild(tr);});return t;}
-function fill(id,p,build){var s=document.getElementById("p-"+id);var b=s.querySelector(".body");var h2=s.querySelector("h2");
- while(h2.childNodes.length>1)h2.removeChild(h2.lastChild);h2.appendChild(badge(p.status));
- var nb=h("div",{"class":"body"});try{build(nb,p.data||{},p);}catch(e){nb.appendChild(h("div",{"class":"err"},"render error: "+e));}
- nb.appendChild(err(p));nb.appendChild(src(p));s.replaceChild(nb,b);}
-
-var R={};
-R.pulse=function(b,d){
- var v=d.violations||[];
- if(v.length){var band=h("div",{"class":"band"});band.appendChild(h("div",{"class":"top"},v[0]));
-  v.slice(1).forEach(function(x){band.appendChild(h("div",null,x));});b.appendChild(band);}
- else if(d.exit_code===0)b.appendChild(h("div",{"class":"OK"},"No violations."));
- var m=d.metrics||{};b.appendChild(table(Object.keys(m).map(function(k){return [k,m[k]];})));
- if((d.unknown_metrics||[]).length)b.appendChild(h("p",{"class":"mute"},"UNKNOWN metrics: "+d.unknown_metrics.join(", ")));
- (d.stale||[]).forEach(function(x){b.appendChild(h("p",{"class":"STALE"},x));});};
-R.agents=function(b,d){var c=h("div",{"class":"cards"});
- (d.seats||[]).forEach(function(s){var card=h("div",{"class":"card"+(s.blocked?" blocked":"")});
-  card.appendChild(h("div",null,h("b",null,s.slug)," ",badge(s.state)));
-  if(s.state==="UNKNOWN"){card.appendChild(h("div",{"class":"mute"},s.error||"no status file"));}
-  else{card.appendChild(h("div",null,"model: "+show(s.model)));card.appendChild(h("div",null,"doing: "+show(s.doing)));
-   card.appendChild(h("div",null,"slice: "+show(s.slice)));card.appendChild(h("div",null,"updated: "+show(s.updated_min_ago)+" min ago"));
-   if(s.blocked_on)card.appendChild(h("div",{"class":"VIOLATION"},"blocked on: "+s.blocked_on));}
-  var tk=s.tick||{};card.appendChild(h("div",null,"last tick: "+show(tk.last_tick)+" ("+show(tk.outcome)+")"));
-  card.appendChild(h("div",null,"next due: "+show(tk.next_due)+(tk.state==="STALE"?" STALE (no tick in 2x "+tk.interval_min+" min)":"")));
-  (s.root_causes||[]).forEach(function(r){card.appendChild(h("div",{"class":"mute"},r));});c.appendChild(card);});
- b.appendChild(c);};
-R.board=function(b,d){var f=h("div",{"class":"flow"});var cs=d.counts||{};
- ["ready","building","review","approved","merged","released"].forEach(function(t){f.appendChild(h("div",null,h("b",null,show(cs[t])),t));});b.appendChild(f);
- if(d.ready_below_30)b.appendChild(h("div",{"class":"err"},"ready queue below "+d.ready_target+" (D91)"));
- b.appendChild(table((d.active||[]).map(function(r){return [r.id,r.owner,r.tier,r.token,show(r.age_min)+" / "+show(r.budget_min)+" min"+(r.over_budget?" OVER":"")];}),["ID","Owner","Tier","Status","Age / budget"]));
- if((d.unparsed||[]).length)b.appendChild(h("p",{"class":"mute"},"Unparsed rows: "+d.unparsed.join(", ")));};
-R.releases=function(b,d){
- b.appendChild(table([["npm latest",show(d.latest)],["npm next",show(d.next)],["VERSION",show(d.version_file)],["last tag",show(d.last_tag)],
-  ["minutes since last publish",show(d.last_publish_minutes_ago)],["releases last hour",show(d.releases_last_hour)],["releases last 24h",show(d.releases_last_24h)]]));
- b.appendChild(h("p",{"class":"mute"},d.npm_label));
- (d.flags||[]).forEach(function(x){b.appendChild(h("div",{"class":"VIOLATION"},x));});
- b.appendChild(table((d.runs||[]).slice(0,8).map(function(r){var a=h("a",{href:r.url||"#",rel:"noreferrer",target:"_blank"},String(r.databaseId));return [a,show(r.conclusion||r.status),(r.headSha||"").slice(0,8)];}),["Run","Result","SHA"]));
- b.appendChild(table((d.versions||[]).slice(0,10).map(function(v){return [v.version,v.published];}),["Version","Published"]));};
-R.okrs=function(b,d){b.appendChild(table((d.items||[]).map(function(i){
- var cur=i.current?(show(i.current.value)+(i.current.target!==undefined?" / "+i.current.target:"")):"";
- return [i.n,i.title,show(i.owner),h("span",{"class":i.state},i.state),cur];}),["#","Item","Owner","State","Value / target"]));};
-R.roi=function(b,d){var c=d.cost||{},v=d.value||{},r=d.roi||{},m=c.metered||{},u=c.usage||{};
- b.appendChild(h("h3",null,"Agent cost ("+c.label+")"));
- b.appendChild(table([["plan USD per month",show(c.plan_usd_per_month)],["plan USD per day (ESTIMATE)",show(c.plan_usd_per_day)],["daily budget USD",show(c.daily_budget_usd)],
-  ["weekly plan-usage percent",show(u.weekly_percent)],["5h window percent",show(u.window_percent)],
-  ["metered run cost",m.state==="MEASURED"?("USD "+m.total_usd+" over "+m.measured_runs+" runs ("+m.unmeasured_runs+" unmeasured)"):("UNMEASURED: "+show(m.reason))]]));
- b.appendChild(h("p",{"class":"mute"},c.label_note+". Usage numbers are plan percentages, not tokens or dollars."));
- b.appendChild(h("h3",null,"Human-equivalent value ("+show(v.state||"standing payroll only")+")"));
- var pr=v.standing_payroll_equivalent;
- b.appendChild(table([["standing payroll equivalent per day",pr?("USD "+pr.per_day_usd+" for "+pr.seats+" seats, "+pr.label):"UNKNOWN"],
-  ["human-equivalent value",v.human_equivalent_usd===null||v.human_equivalent_usd===undefined?"UNMEASURED":"USD "+v.human_equivalent_usd],["ROI multiple",r.multiple===null?"not printed ("+r.note+")":r.multiple+"x ("+r.weaker_side_label+")"],
-  ["cost per verified outcome",(d.scoreboard||{}).cost_per_verified_outcome_usd===null?"UNMEASURED":"USD "+d.scoreboard.cost_per_verified_outcome_usd]]));
- b.appendChild(h("p",{"class":"mute"},"Formula: "+v.formula));
- var bs=d.basis||{};b.appendChild(h("p",{"class":"mute"},"Basis: "+bs.path+" at "+show(bs.git_sha)+", snapshot "+show(bs.snapshot_date)+(bs.stale?" (STALE: "+bs.stale_note+")":"")+", source "+show(bs.source_url)));
- (d.notes||[]).forEach(function(x){b.appendChild(h("div",{"class":"err"},x));});};
-R.queue=function(b,d){b.appendChild(h("p",null,"Rows: "+show(d.rows_total)+", last number "+show(d.max_number)+", from cockpit "+show(d.cockpit_rows)));
- b.appendChild(table((d.last||[]).map(function(r){return [r.n,r.date,r.item];}),["#","Date","Item"]));};
-
-function render(snap){var ps=snap.panels||{};Object.keys(R).forEach(function(k){if(ps[k])fill(k,ps[k],R[k]);});
- document.getElementById("refreshed").textContent="last refreshed "+new Date().toLocaleTimeString()+" (generated "+snap.generated_at+")";
- document.getElementById("fail").textContent="";}
-function poll(){fetch("/api/snapshot",{cache:"no-store"}).then(function(r){return r.json();}).then(render)
- .catch(function(e){document.getElementById("fail").textContent="refresh failed: "+e;});}
-poll();setInterval(poll,15000);
-
-var item=document.getElementById("item");
-item.addEventListener("input",function(){document.getElementById("count").textContent=item.value.length+"/2000";});
-document.getElementById("send").addEventListener("click",function(){
- var btn=this,res=document.getElementById("qres");btn.disabled=true;res.textContent="sending";
- fetch("/api/queue",{method:"POST",headers:{"Content-Type":"application/json","X-Cockpit-Token":TOKEN},
-  body:JSON.stringify({item:item.value,why:document.getElementById("why").value})})
- .then(function(r){return r.json().then(function(j){return {ok:r.ok,j:j};});})
- .then(function(x){res.textContent=x.ok?("Appended row "+x.j.row+": "+x.j.line):("Rejected: "+x.j.error);if(x.ok){item.value="";poll();}})
- .catch(function(e){res.textContent="Failed: "+e;}).then(function(){btn.disabled=false;});});
-})();
-</script>
-</body>
-</html>

package/tools/cockpit/roi-basis.json

@@ -1,33 +0,0 @@
-{
-  "source_url": "https://www.levels.fyi/companies/anthropic/salaries/software-engineer",
-  "snapshot_date": "2026-09-24",
-  "refetched_on": "2026-10-23",
-  "annual_total_comp_usd": {
-    "entry": 483000,
-    "senior": 621000,
-    "lead": 865000,
-    "staff": 1410000,
-    "median": 883000
-  },
-  "research_scientist_median_usd": {"value": null, "status": "UNVERIFIED"},
-  "research_engineer_usd": {"value": null, "status": "NOT FOUND"},
-  "overhead_multiplier": {"value": 1.25, "status": "ASSUMPTION"},
-  "work_hours_per_year": {"value": 2000, "status": "ASSUMPTION"},
-  "default_tier": "senior",
-  "seat_tier_status": "ASSUMPTION",
-  "seat_tier": {
-    "chief-of-staff": "lead",
-    "cto": "staff",
-    "developer-experience-lead": "lead",
-    "platform-lead": "lead",
-    "product-growth-lead": "lead",
-    "release-reliability-lead": "lead",
-    "research-harness-scientist": "lead",
-    "research-outcomes-scientist": "lead",
-    "security-lead": "lead",
-    "verification-scientist": "lead"
-  },
-  "hours_per_outcome": {"verified_slice": {"value": null, "status": "ASSUMPTION, unset"}},
-  "plan_usd_per_month": 200,
-  "daily_budget_usd": 50
-}