Skip to content

donaldfilimon/abi / .agents/skills/goals

Capture, track, and execute goals for the abi stable-Rust project. Use when the user wants to record a goal, see outstanding goals, update goal status, or break a goal into a verified work slice. Distinct from abi-goal-orchestrator (which plans from the TODO board) — this skill owns a persistent goal ledger and drives goal-to-slice execution.

Goals

A lightweight goal ledger for the abi project plus a loop that turns a goal into a verifiable work slice and records the outcome. Goals persist in tasks/goals.md so they survive sessions; the active board (tasks/todo.md) remains the granular working list.

Ledger

The ledger is tasks/goals.md, a markdown file with one ## <Goal> section per goal. Each section carries a status line and a short body:

## Ship file-aware agent context
status: done
- Wired @file resolution into agent plan/multi + TUI repl

Valid status: values: todo, in_progress, blocked, done.

If tasks/goals.md does not exist, create it with a # Goals header. Treat the file as the source of truth; never delete a closed goal — move a one-line note to the bottom or mark done.

Workflow

  1. Start in the abi repo root unless the user gives another checkout.
  2. Inspect git status --short --branch before edits; preserve unrelated dirty work.
  3. Read AGENTS.md, tasks/todo.md, and tasks/lessons.md for conventions.
  4. Capture: if the user states a goal, ensure it has a ## <Goal> section in tasks/goals.md with status: todo (or in_progress if started this turn).
  5. Track: list goals by reading tasks/goals.md. update <goal> <status> rewrites that goal's status: line.
  6. Execute: derive the smallest slice that moves the goal forward. Prefer a change that makes one TODO, roadmap gap, doc mismatch, or validation gap measurably more true. Implement it, then verify.
  7. Keep claims honest: source/build/tests override prose. Update the goal's status and append a one-line outcome note when the slice lands.

Goal Rules

  • Do not convert disclosed stubs into fake completions. Native GPU kernels, production clustering, production FHE, and learned-compression claims need real source/tests.
  • Do not add legacy CLI names. Preserve the frozen top-level command set and the MCP 12-tool contract.
  • When changing docs, keep claims aligned with docs/contracts/external-claims-audit.mdx.
  • Goals are user intentions, not tasks — keep tasks/goals.md coarse and let tasks/todo.md hold the granular steps.

Verification Ladder

Prove the slice with the narrow command first, then broaden:

./tools/cargo.sh test -p <crate> --lib -- <filter>
./tools/check.sh

Smoke the real binaries after surface changes:

./tools/cargo.sh build -p abi-cli -p abi-mcp
./target/debug/abi backends
./target/debug/abi complete "hi"
printf '%s\n' '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | ./target/debug/abi-mcp 2>/dev/null

Always use ./tools/cargo.sh — never bare Homebrew cargo.

Notes

  • This skill is project-scoped (lives under .agents/skills/goals/); it does not change the CLI surface or MCP tools.
  • Mirrors the taxonomy and honesty constraints of abi-goal-orchestrator but owns the persistent ledger rather than reading the TODO board only.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall