jedbjorn
UserForkable shell substrate for a single code repository — DB-backed identity, memory, roadmap, content; harness-agnostic boot. We build the data layer, we rent the harness.
Categories
Indexed Skills (41)
cartographer
Own the repo map — configure mapping to THIS repo, wire the auto-remap git hooks, heal both on drift. Cartographer-only; no working shell maps. Run on first boot + whenever the map looks wrong.
self_update
Update this fork's super-coder engine in place — fetch + materialize new code + migrations, all memory intact; sound rollback. The shell hands off to its own next boot. Use when a super-coder update is available.
snapshot
Serialize DB work via sc snapshot / sc render under the instance artifact policy. Tracked mode publishes through Git; local mode persists under .sc-state/local without creating content commits.
docs
Author or review docs & specs in super-coder. The DB owns the body (documents table); roadmap tracks specs (the dev cycle), the Docs tab holds docs. Use whenever asked for a doc, spec, report, design, RFC, ADR, runbook, or to edit existing ones.
flag_sweep
Admin's every-session flag reconciliation — auto-close flags whose gating work is provably done, open ship flags for implemented-but-unshipped specs and docs-pending flags for shipped features that lack a doc (message the planner), surface judgment calls to the FnB. Step 1 of the admin standing pass; run before git_cleanup.
memory
When + how this shell persists memory — current_state (≤300), session narrative, seed (cap 10), L&S (cap 20, ≤500/entry, --supersedes|--new), decisions — all via sc mem, written as it happens, not at close.
spec
Execute a spec across sessions — analyze viability, surface blockers and unclear items, break into tasks (Preparation → impl steps → Verification), and track progress in spec_tasks. Updates current_state at every step. Load when starting, implementing, or building any feature, spec, or roadmap item — before writing code.
curate
The periodic L&S sweep. Run when the STATUS L&S line says "curation due" — resolve contradictions, merge entries stating one rule, promote a recurring process to a skill, move environment facts out, then stamp `sc mem curated`. Yours alone; never delegate it.
local_skill_management
Create, persist, assign, and remove fork-specific skills — the correct authoring path so skills survive snapshot/rebuild cycles.
engine_surgery
Procedure for changing the engine you are running — pull/reconcile/restart cadence and their costs, three-artifact engine-skill commits with a hermetic mirror render, migrating the live DB safely, and verifying claims about engine code against the remote rather than a possibly-stale checkout. SOURCE-REPO ONLY; a fork consumes the engine as a pinned dependency and never edits it. Load before touching .super-coder/ in the repo that owns it.
db_map
Data model behind the engine memory surfaces + the `sc mem` command for each. Check before reading or writing memory — identity, decisions, roadmap, documents, flags. Reads/writes go through the API (`sc mem`), never raw sqlite.
sprint_orchestration
Run the steady-state planner loop for a multi-shell sprint. Declare the sprint and structured unit board, assign developers and reviewers, arm event-driven wake, dispatch scoped tasks, advance units from durable events, and route merge order. Load sprint_orchestration_recover only when an expected transition stalls or an alert opens. Load sprint_orchestration_close only when every unit is terminal and main is green.
sprint_dev
Execute a developer unit in an ACTIVE structured sprint. Read the assigned unit, build within its scope, persist long work through sc job, open and register the PR, drive CI and review, merge the exact approved head under scoped authority, and send a structured unit report. Load when the boot sprint directive or a scoped task assigns a developer unit.
sprint_orchestration_close
Close a sprint after every unit is terminal and main is green. Run an independent spec-to-main conformance pass, route findings while authority remains active, close and freeze the sprint document, verify wake and PR-watch cleanup, synthesize the report, and settle flags and roadmap state. Load only at the close trigger.
sprint_orchestration_recover
Recover a declared sprint after a critical wake alert or a missing expected transition. Diagnose from the board, binding, messages, jobs, PR state, and worker process evidence; preserve completed work; apply one bounded repair; return to sprint_orchestration. Load only after the steady-state loop identifies a stall.
sprint_review
Execute an assigned sprint unit review or the close-time conformance pass. Review the requested exact head, test high-value properties adversarially, return Major/Medium/Low findings and a recommendation to the planner, declare review-clean explicitly, or write requirement-level conformance verdicts against integrated main. Load when the boot sprint directive or a scoped task assigns review work.
issue_reporting
Report engine defects upstream — the moment a ./sc command fails or lies, a skill contradicts your reality, the API blocks a documented workflow, or you work around the engine to proceed. File a GitHub issue on super-coder; your repo's app bugs stay in the fork.
git
Git conventions for a super-coder shell — one repo, one cwd. Sync the base before work, branch before committing, open PRs (never merge without the FnB's OK), attribute commits per-shell. Use before any git work.
dev_kit
What the sandbox dev kit provides + how to drive it — ./sc deps, ./sc test, ./sc lint, ./sc typecheck, the .venv tools, rg/sqlite3, the baked browser, the container/host app boundary, and the optional app-only Postgres sidecar (DATABASE_URL). Use when building or testing in a fork.
agents
--agents [model] — delegate work to spawned subagents under the system's discipline. Dev — execute a spec's task plan as implementer waves; reviewer — fan the three review axes out to an adversarial finding-panel. Overlay on spec/review; parent-only memory writes; AGENTS spawn ledger with a hard 6h validity window; parent-set timeouts. Load ONLY when the FnB invokes --agents.
app_deploy_setup
Admin-run, one-time scaffold — turn the shipped deploy template into this repo's own project-local `deploy` skill (migration dirs, DB backup, ff-only sync, apply + move migrations, restart), then grant it to every shell.
bootstrap
First-run orientation. Run ONCE when the boot doc shows "## FIRST RUN" (bootstrapped=0), BEFORE other work — read the repo map + your identity, set current_state, mark yourself oriented.
configure_winbox
Provision the Windows Test VM — push the fork's committed winget manifest to the guest via the vm-broker, install + verify what the MANIFEST says, then hand the operator the one-command snapshot bake (./sc vm-bake). Admin-only; runs before the snapshot, re-runs on any toolchain change.
database-migrations
Database migration safety + how super-coder's own migrations work (schema.sql baseline + ordered migrations/ deltas + ledger). Use when altering tables, adding columns, or running backfills — in the host repo's DB or super-coder's.
flags
Track blockers as flags — surface open ones, open new ones, edit long-lived ones, resolve them. Link a flag to the roadmap feature it blocks. Mirrors the GUI Flags tab. Use when something blocks progress or needs follow-up.
migration_management
Author and apply fork-specific DB schema migrations — naming, format, how to apply locally and verify.
onboard
One-time, FnB-supervised ingest of a repo's EXISTING docs/specs into the DB + roadmap backfill — the only time content flows file→DB. Run once after bootstrap on a fork with existing documentation. Planning shell's job.
pm2
Observe + manage the host's pm2-supervised app stack through the host-side pm2-broker — status, app health, log tails, and scoped restarts, holding no host access yourself. The admin shell's deploy-confirmation companion. Use when verifying a deploy, checking the live app, or bouncing a declared process.
query_authoring_pg
Compose + run SQL against a Postgres-backed fork's app DB — psql mechanics, psql variables vs driver :params, SQLite→Postgres dialect traps, read-only diagnostics, paste-ready handoff when the DB is outside the sandbox. Use when diagnosing app data issues or verifying data by query.
review
Reviewer procedure — read a diff against its spec along three axes (code quality, edge cases & gaps, spec conformance), open flags for failures, then propose the handoff (fixes to dev / new spec to planner) to the FnB and send it only on approval. The reviewer's top-level loop; the lenses live in the skills it points to. Load when reviewing a dev's work.
surface_catalogue
Read the host repo via the dr_* catalogue (files, languages, deps, env) BEFORE grepping or walking the tree. Query first, lazy-load the few files it points at. Use to orient in an unfamiliar repo fast.
tailscale
Reach the fork's hosts over the tailnet — read tailnet status and exec commands on a tailnet host through the host-side ts-broker, holding no tailnet credential yourself. The devops shell's signature skill. Use when operating remote hosts, deploys, or backups over Tailscale.
test_authoring
Principles for stringent pytest tests — tests a realistic bug turns red. Pair with a granted stack-infra testing skill (test_authoring_sqlite / test_authoring_pg / a fork-local one) if the shell has one.
test_authoring_pg
Postgres test infrastructure for postgres-backed forks — throwaway DB, Alice/Bob tenants, psycopg 3 direct assertions. Read alongside test_authoring for the rules.
test_authoring_sqlite
SQLite test infrastructure for super-coder-style forks — throwaway DB, Alice/Bob tenants, Caller/TestClient. Read alongside test_authoring for the rules.
windows_devkit
Drive the linked Windows Test VM — push a build artifact, exec the installer/test over SSH, capture output + a screenshot, then reset to the clean snapshot. High-fidelity installer/system-level testing where Wine is useless. Use when building or verifying Windows software in a fork that has a configured VM.
windows_vm_gui
Drive the Windows Test VM's GUI — Windows-MCP in the guest, UIA-tree clicking by element ID (never blind coordinates), screenshot verification between actions, and mouse-free in-process test paths for anything repeatable. Exploratory GUI QAQC where windows_devkit's exec loop can't see the UI.
api-design
REST/HTTP API design patterns — resource naming, status codes, pagination, filtering, errors, versioning, idempotency. Use when designing or reviewing API endpoints.
authoring_syntax
House syntax for text authored for AI consumption — skill bodies, shell focus/mandate text, agent prompts. Directives with success conditions, specific prohibitions, light operators, need-to-know. Fires when authoring or editing one, or on "syntax check X".
blueprint
Turn a one-line objective into a sequenced construction plan — decompose into steps, find the dependency order, mark what can run in parallel, name the verification gate. Use before multi-step builds.
redline_review
Review PNG redlines from the shared scratch dir — find by filename match, describe what is seen, interpret intent, propose implementation, hold for approval before any code. Fires when the FnB says "redlines".
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.