10ximplement is a Codex skill for long-running implementation work that needs more structure than a single prompt.
It combines four habits into one workflow:
- keep a durable implementation ledger
- choose narrow PM-style phase gates
- execute plans with subagents and review loops when available
- fall back to sequential plan execution when subagents are not the right fit
The key artifact is implementation-notes.html: a readable run ledger that tracks resume state, phase gates, decisions and the logic behind them, validation, blockers, accepted defers, evidence, and the next exact action.
Clone this repo directly into your Codex skills folder:
git clone https://github.com/cobibean/10ximplement.git ~/.codex/skills/10ximplementThen invoke it in Codex:
Use $10ximplement to run this build through phase gates, subagent execution, review checkpoints, and a durable implementation ledger.
10ximplement is built for work where agents usually lose the plot:
- multi-phase builds
- plans that need review gates
- subagent-heavy implementation
- long context windows and compaction risk
- decisions that need to survive handoff
- "continue where we left off" sessions
The skill creates this structure inside the target project:
.agent/
GOALS.md
runs/
<goal-id>/
GOAL.md
implementation-notes.html
evidence/
GOAL.md is the contract. implementation-notes.html is the live state and history. evidence/ stores bulky logs, screenshots, reports, or command output that should not clog the HTML.
The skill ships a helper script:
python3 ~/.codex/skills/10ximplement/scripts/init_10x_ledger.py \
--root . \
--goal-id "my-feature" \
--title "My Feature" \
--objective "Ship the feature through reviewed phase gates." \
--mode fullThe script uses assets/implementation-notes.template.html, so agents do not need to recreate the ledger UI from memory.
At a high level, the agent:
- Loads repo/spec context and checks current worktree state.
- Creates or resumes the
.agent/runs/<goal-id>/ledger. - Chooses one narrow implementation phase.
- Asks for PM/review mode at phase boundaries when needed.
- Executes through subagents, sequential plan steps, or a PM handoff.
- Runs review gates before accepting the checkpoint.
- Updates
implementation-notes.htmlwith decisions, logic, validation, defers, blockers, and next action. - Finishes only after checking
GOAL.mdfinishing criteria.
This skill synthesizes ideas from:
- Goal Ledger: local custom skill for
.agent/runs/<goal-id>/GOAL.mdandimplementation-notes.html - Product Manager Phase Gates: local custom skill for PM modes, review modes, narrow slices, and checkpoint verdicts
- executing-plans
- subagent-driven-development
It also inherits useful surrounding practices from the Superpowers ecosystem:
SKILL.md: main skill instructionsassets/implementation-notes.template.html: run ledger templatescripts/init_10x_ledger.py: ledger initializerreferences/subagent-prompts.md: compact subagent prompt guidanceagents/openai.yaml: Codex UI metadata
MIT