build
Take a refined ticket from spec through plan, implement, review, and complete.
Skill by charleshall888
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/charleshall888-build
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: research
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 26, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/build
What's inside
- Skill name
build- Description
- Take a refined ticket from spec through plan, implement, review, and complete. Needs research.md and spec.md — run refine first if either is missing.
- License
- MIT
- Instructions
- 93 lines in SKILL.md · view on GitHub
Files (16)
- SKILL.md
- lifecycle.config.md · assets/
- backlog-writeback.md · references/
- competing-plans.md · references/
- complete-first-run.md · references/
- complete.md · references/
- implement.md · references/
- kept-pauses-data.toml · references/
- orchestrator-review.md · references/
- parallel-execution.md · references/
- plan.md · references/
- protocol-expectation.txt · references/
- review.md · references/
- size-pin.txt · references/
- wontfix.md · references/
- worktree-entry.md · references/
SKILL.md (first lines)
# Build
Plan → implement → review → complete. State lives in files, so it survives context loss. `cortex/lifecycle.config.md`, when present, overrides complexity defaults, test commands, phase skipping, and review criteria.
## Step 1: Read the served state
```bash
cortex-lifecycle-next "$ARGUMENTS" --expect-file ${CLAUDE_SKILL_DIR}/references/protocol-expectation.txt
```
One read-only call returns the state, its advance contract, and its pause spec. `--expect-file` lets it detect a version mismatch between the installed CLI and this skill. Use the JSON result's fields, not the legacy `next` field. **Stop on `protocol-skew`, wrapper exit 2 (CLI not installed), or a missing command** — each gives its own fix; show it and stop.
Forms: `/cortex-core:build <feature>`, `<feature> <phase>`, reserved `complete <slug>` / `resume <feature>`.
**Not yet refined** — state `research`, `specify`, or `new`: hand off to `/cortex-core:refine {feature}`. Never plan without both `research.md` and `spec.md`; overnight needs both.
**`resume`** is served by phase: `state` is the current phase, `advance_contract` goes to `cortex-lifecycle-advance` at each boundary, and `pause_spec` drives the kept pauses. Write artifacts under `roots.artifacts.path` — never the log path, which always points at the main checkout.
<!-- pause: empty-lifecycle-offer question -->
<!-- pause: ambiguous-backlog-pick question -->
**Other states** carry a `next` instruction — act on it:
- `derive-slug` → make a 3–6 word kebab slug and re-run without asking.
- `empty` → offer the incomplete `cortex/lifecycle/*` lifecycles, then re-run.
- `ambiguous-backlog` → present `candidates`, then re-run.
- `wontfix` → run the named `cortex-lifecycle-wontfix` command and stop.
- `closed` / `parked` → already decided; show `next`, do not build.
- `error` / `needs-feature` / `no-such-lifecycle` → report and stop.
## Step 2: Enter the resolved state
Run the result's `enter_command` exactly as given — a `cortex-lifecycle-enter` call with every argument already filled in:
```bash
{envelope.enter_command}
```
Never rebuild it or swap the user's typed token in for `--feature`. Its backlog-file argument lets a resume repair an index that never got its backlog tags.
- `ready` → proceed.
- `needs-decision` → the item is `already_complete`; nothing ran. Apply the Backlog Status Check in [backlog-writeback.md](${CLAUDE_SKILL_DIR}/references/backlog-writeback.md).
- `blocked` → a check refused, and the user can fix the cause; `.session` is unwritten. Stop, fix, re-run.
- `ensure-failed` / `error` → stop.
- Exit 2 → ambiguous slug; follow backlog-writeback.md's exit-2 rule.
Mention any `ignored_tokens` in one line.
On resume, report the served `state`/`criticality`/`tier`, offer continue (default) or restart, and mention `staleness` briefly. **Carry `criticality` and `tier` forward** — the phase references use them instead of re-reading.
## Step 3: Execute the phase
| Phase | Reference | Artifact |
|-------|-----------|----------|
| Plan | [plan.md](${CLAUDE_SKILL_DIR}/references/plan.md) | `plan.md` |
| Implement | [implement.md](${CLAUDE_SKILL_DIR}/references/implement.md) | Source + commits |
| Review | [review.md](${CLAUDE_SKILL_DIR}/references/review.md) | `review.md` |
| Complete | [complete.md](${CLAUDE_SKILL_DIR}/references/complete.md) | Git workflow + summary |
Read **only** the row for the served `state`.
Invoking this skill authorises the sub-agents a phase reference calls for, even under a standing "no agents unless asked" rule. If you cannot start them, say what you did instead in the phase summary and label any review a self-review.
**Path propagation.** `${CLAUDE_SKILL_DIR}` resolves only in this body. Wherever a reference names a `${CLAUDE_SKILL_DIR}/…` path, use the absolute path resolved here.
## Advance-verb routing
Every phase boundary hands off to `cortex-lifecycle-advance`. It writes that boundary's events in order and is safe to run again. Route on the returned `state` and show its `message` / `reason` / `preferred_remedy`. Never work out an outcome or write an event by hand. On `refused`, re-run `cortex-lifecycle-next`, then re-invoke with `--from-state` set to `advance_contract.expected_from_state`; if still refused, escalate, naming both phases. Verb missing from `PATH` → stop; the operator installs or upgrades the cortex-command CLI.
## Phase transitions
Cross boundaries automatically: announce and continue. Add no stop unless a `<!-- pause: -->` marker or the routed outcome says so. Each summary carries **Decisions**, **Scope delta**, **Blockers** ("None" when empty), then **Next**. A boundary fires on its condition (e.g. `plan.md` all tasks `[x]`), not on user input; Plan also needs user approval.
## Criticality
Override anytime: `cortex-lifecycle-event criticality-override --feature <name> --from <old> --to <new> --reason "{tag}: <one line>"`. Tag: `reversibility:` / `exposure:` / `consequence:` / `other:`. The reason saves the next reader from working it out again.
`cortex-lifecycle-state --feature {feature}` (or `--field <x>`) returns the log's current values and omits absent keys — default `criticality=medium` / `tier=moderate` yourself. **`"corrupted": true`** → tier and criticality are unknown: treat the feature as *requiring* review.
- **Review** runs at every tier when criticality is `high`/`critical` (Stage 2 complex-only). Otherwise it runs for complex tier only.
- **Orchestrator review** runs at every phase boundary, except `low` criticality below complex.
- **Planning** uses competing plans at `critical`, a single plan otherwise.
You choose the model each time you start an agent. The implement→{review|complete} routing lives in its verb.
## Situational references
- [parallel-execution.md](${CLAUDE_SKILL_DIR}/references/parallel-execution.md) — parallel features via `Agent(isolation: "worktree")`
- [wontfix.md](${CLAUDE_SKILL_DIR}/references/wontfix.md) — operator-decided termination
<!-- pause: resume-feature-pick question -->
A session binds to one feature through `cortex/lifecycle/{feature}/.session` — gitignored, cleaned at SessionEnd, never committed. Several incomplete lifecycles and none named → list them and ask. Ignore features with `feature_complete` in events.log or an APPROVED verdict in review.md.From charleshall888/cortex-command, MIT.
More from charleshall888/cortex-command
- 0.606score
overnight
charleshall888
Plan and launch an unattended overnight session — pick refined features, approve the plan, hand off to the runner.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
android-cli
charleshall888
Orchestrates Android development tasks including project creation, deployment, SDK management, and environment diagnostics using the `android` command-line…
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
critical-review
charleshall888
Adversarial review — 1–2 fresh reviewer agents on distinct angles, consolidated here.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
dev
charleshall888
Development entry point — routes a request to the right workflow.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
discovery
charleshall888
Investigate a topic, then decompose it into backlog tickets grouped by epic.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
refine
charleshall888
Take a backlog item from idea to approved spec via Clarify → Research → Spec.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
research
charleshall888
Parallel research orchestrator — dispatches 1–6 agents across independent angles, synthesizes into research.md.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
morning-review
charleshall888
Walk the morning report after an overnight run — verify features, answer deferred questions, advance lifecycles, merge the PR, close tickets.
SkillAFreshness: A★ 3Claude Code +9 - 0.604score
harness-review
charleshall888
Evaluate overnight runner components against a curated inventory using a 4-question rubric to classify each as load-bearing, experiment-candidate, or…
SkillAFreshness: A★ 3Claude Code +9 - 0.602score
r8-analyzer
Google LLC
Analyzes Android build files and R8 keep rules to identify redundancies, broad package-wide rules, and rules that subsume library consumer keep rules.
SkillAFreshness: A★ 3Claude Code +9 - 0.601score
requirements
charleshall888
Use /cortex-core:requirements to gather requirements or define project scope.
SkillAFreshness: A★ 3Claude Code +9 - 0.601score
writing-great-skills
charleshall888
Reference for writing and editing skills well — the vocabulary and principles that make a skill predictable.
SkillAFreshness: A★ 3Claude Code +9
Signals
- Stars
- 3
- Forks
- 0
- Open issues
- 0
- Last commit (this folder)
- Sep 22, 2026 (1 month ago)
- Last release
- Oct 5, 2026 (24 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/charleshall888/cortex-command
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from charleshall888
- 0.606score
refine
charleshall888
Take a backlog item from idea to approved spec via Clarify → Research → Spec.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
discovery
charleshall888
Investigate a topic, then decompose it into backlog tickets grouped by epic.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
android-cli
charleshall888
Orchestrates Android development tasks including project creation, deployment, SDK management, and environment diagnostics using the `android` command-line…
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
dev
charleshall888
Development entry point — routes a request to the right workflow.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
critical-review
charleshall888
Adversarial review — 1–2 fresh reviewer agents on distinct angles, consolidated here.
SkillAFreshness: A★ 3Claude Code +9 - 0.606score
research
charleshall888
Parallel research orchestrator — dispatches 1–6 agents across independent angles, synthesizes into research.md.
SkillAFreshness: A★ 3Claude Code +9
Cite this page
build skill. Agent Mart. https://agentmart.dev/skills/charleshall888-build (last updated 2026-10-24)