Skip to content
Back to skills

Decisions

ASecurity

**Decision.** The repository ships one coding subagent, `plan-implementer`, in each client's own project format — `.claude/agents/plan-implementer.md`, `.codex/agents/plan-implementer.toml`, `.kiro/agents/plan-implementer.json` — with the same text. That text does not restate the procedure: it points at the "Canonical mandate" in `.agents/skills/implement-issue/SKILL.md` §2 and fixes only the defaults a coordinator would otherwise repeat in every mandate (stay in the worktree, never merge, st...

  • 55 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added October 14, 2026
ai-agentsgo

Works with

  • cli

Security analysis

A100/100

Pro scans all 21 files and shows the line behind each finding

Scanned October 14, 2026

npx -y skills add BeppeTemp/cartographer --skill decisions --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Decisions?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Decisions
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/beppetemp-decisions/badge)](https://www.skillsdirectory.com/skills/beppetemp-decisions)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
topic: project-governance
---

# D327 — The plan-implementer subagent is a thin per-client wrapper over the implement-issue skill

**Decision.** The repository ships one coding subagent, `plan-implementer`, in
each client's own project format — `.claude/agents/plan-implementer.md`,
`.codex/agents/plan-implementer.toml`, `.kiro/agents/plan-implementer.json` —
with the same text. That text does not restate the procedure: it points at the
"Canonical mandate" in `.agents/skills/implement-issue/SKILL.md` §2 and fixes
only the defaults a coordinator would otherwise repeat in every mandate (stay in
the worktree, never merge, stop on a stale plan, a ≤10-line report).

**Why.** Landing a batch of plans spawns one subagent per plan; with the mandate
inline, every spawn carries the same long prompt and every report comes back in
a different shape. A named agent shrinks the mandate to "issue #N, worktree X"
and makes the report predictable. The cost is three files that must say the same
thing, because no client reads another's agent format and a symlink cannot
translate between Markdown, TOML and JSON.

**Alternatives rejected.**
- Duplicating the mandate in each agent file: three copies of a procedure drift;
  the skill is already the single copy every client reads (D203).
- A Cartographer-provisioned agent: provisioning serves KB artifacts to a user's
  machine, not a repository's own contributor tooling.
- No agent, inline mandates only: works on every client, but costs the prompt on
  every spawn and leaves the report format to chance.

**Model.** Each definition pins a cheaper model than a coordinator usually runs on (Claude `claude-sonnet-5-5`, Kiro `claude-sonnet-5.5`, Codex `terra-6`): a plan issue already carries the design, the implementer executes it, and the coordinator's diff read is the quality gate. A model rename is a one-line change in each of the three files.

**Consequences.** A change to the mandate goes in the skill only. A change to the
wrapper's defaults goes in all three files in the same commit. Antigravity has no
documented project-local agent cell, so it keeps running the mandate inline.

Files in this skill

  • AD1-AD13-original-architecture.md2.8 KB
  • D1-stdlib-only-constraint-removed.md330 B
  • D10-concept-write-frontmatter-from-json-map.md367 B
  • D100-community-surfaces-pages-from-docs-discussions-on-wiki.md1.6 KB
  • D101-tool-names-in-the-docs-are-ci-enforced.md1.9 KB
  • D102-opt-in-per-kb-mcp-tool-name-prefix.md2.8 KB
  • D103-observable-replication-failures-and-native-git.md758 B
  • D104-structured-sops-pointers-scoped-refs-and-encrypted.md692 B
  • D105-binary-safe-provisioning-artifacts-and-executable-kb.md992 B
  • D106-concept-owned-non-markdown-assets.md1 KB
  • D107-declarative-lint-contracts-in-map-descriptors.md1.3 KB
  • D108-on-demand-backlinks-and-frontmatter-facets.md896 B
  • D109-kb-owned-concept-templates-and-one-shot-scaffolding.md1.6 KB
  • D110-topic-owned-decision-records-and-github-owned-project.md1.5 KB
  • D111-repository-e2e-tests-are-deterministic-and-model-free.md1.4 KB
  • D112-reserved-local-endpoint-defaults.md1 KB
  • D113-one-client-status-snapshot-across-cli-and-dashboard.md977 B
  • D114-verify-provisioning-artifacts-cryptographically.md1.3 KB
  • D115-mcp-allow-list-and-hash-bound-local-approval.md848 B
  • D116-trusted-stdio-mcp-descriptors-with-environment.md1.6 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.