implement
Implementation through managed, value-routed work packages.
Skill by marcuskrogh
Chalk Labs
Need help deploying implement?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
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/marcuskrogh-implement
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: productivity
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. Checked Oct 25, 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/implement
What's inside
- Skill name
implement- Description
- Implementation through managed, value-routed work packages. Reuses the Task's delivery branch/PR when one exists, honors a Workflow binding when present, and enforces tests and structure as-you-go. Use standalone, for an approved definition spec, or for review fix-forward.
- License
- MIT
- Instructions
- 109 lines in SKILL.md · view on GitHub
Files (4)
- SKILL.md
- rework.md
- structure.md
- testing.md
SKILL.md (first lines)
# Implement Applies [CONCEPT_IMPLEMENTATION](../concepts/CONCEPT_IMPLEMENTATION.md) and [CONCEPT_SKILL](../concepts/CONCEPT_SKILL.md) to the **current repository**. When a Workflow binding is present, honor its parameters — do not reclassify. **On invoke:** read [../concepts/CONCEPT_IMPLEMENTATION.md](../concepts/CONCEPT_IMPLEMENTATION.md), [CONCEPT_SKILL](../concepts/CONCEPT_SKILL.md), [../concepts/CONCEPT_STRUCTURE.md](../concepts/CONCEPT_STRUCTURE.md), [../concepts/STRUCTURE-CATALOG.md](../concepts/STRUCTURE-CATALOG.md), [testing.md](testing.md), [structure.md](structure.md), and [../workflow/SKILL.md](../workflow/SKILL.md). When the definition spec is absent, also read [CONCEPT_ALIGNMENT](../concepts/CONCEPT_ALIGNMENT.md), [CONCEPT_DEFINITION](../concepts/CONCEPT_DEFINITION.md), and [CONCEPT_CLASSIFICATION](../concepts/CONCEPT_CLASSIFICATION.md). When `ARCHITECTURE.md` is absent, also read [CONCEPT_ARCHITECTURE](../concepts/CONCEPT_ARCHITECTURE.md). When `implement.verify` is `comparative` (or the spec is `REWORK.md`), also read [rework.md](rework.md). When `sandbox=inject` or `SANDBOX.md` is present, follow its Promote map (production targets + copy notes). When `implement.mode` is `multiagent` (or spawning workers otherwise), also read [CONCEPT_DELEGATION](../concepts/CONCEPT_DELEGATION.md) and its platform catalog as directed there. ## Extensions | Slot | This skill | |------|------------| | **Spec source** | Tracker Task + Sub-tasks + `PLAN.md` / `BUG.md` / `TWEAK.md` / `REFINE.md` / `REWORK.md` / `ITERATE.md` / `ADOPT.md` / linked specs; load `RESEARCH.md` / `MODEL.md` / `SANDBOX.md` from PLAN Inputs or delivery branch when present | | **Workflow binding** | `PLAN.md` `## Workflow` when present; else legacy fallback in [CLASSIFICATION-CATALOG.md](../concepts/CLASSIFICATION-CATALOG.md#legacy-fallback) | | **Branch naming** | WORKSPACE pattern — **reuse** Task delivery branch if it exists | | **Delivery** | **Same** PR as define/bug/tweak/refine/rework/adopt when one exists (or branch-only per WORKSPACE); research/model/sandbox may have started the branch without a PR | | **Verification** | Per binding `implement.verify`: `tests` → [testing.md](testing.md); `non-regression` → behaviour unchanged + testing.md; `comparative` → [rework.md](rework.md) + testing.md. Plus [structure.md](structure.md) **manager gate**, lint, plan checklist, sub-task completion | | **Testing checklist** | [testing.md](testing.md); comparative adds [rework.md](rework.md) | | **Structure checklist** | [structure.md](structure.md) + [STRUCTURE-CATALOG.md](../concepts/STRUCTURE-CATALOG.md) | | **Closeout gate** | [structure.md](structure.md#manager-gate-before-leaving-implement) + [testing.md](testing.md) on the **whole** diff before this skill's outcome is `built` | | **Model routing** | CONCEPT_DELEGATION when `implement.mode=multiagent` or workers are spawned; `single` → manager may implement localized packages without workers when Routine | | **Work package types** | See table below; add **Promote** when `SANDBOX.md` is promotion-ready | | **PR template** | Summary; Tracker; Spec refs; Workflow binding; Test plan; Structure notes; Completed sub-tasks / review threads | ## Modes | Mode | When | Behavior | |------|------|----------| | **Build** (default) | Task To Do / In Progress | Full implementation loop with tests and structure in-package | | **Fix-forward** | After a laser or code review with must-fix findings; same Task + open PR | Address review threads only; add/adjust tests and structure for the finding axes | ## Spec priority 1. Fix-forward: open PR review comments 2. `PLAN.md` **Workflow** / **Classification** binding (do not override without user ask) 3. Sub-task descriptions 4. Task description 5. `PLAN.md` / `BUG.md` / `TWEAK.md` / `REFINE.md` / `REWORK.md` / `ITERATE.md` / `ADOPT.md` / linked specs 6. `RESEARCH.md` / `MODEL.md` on the delivery branch (PLAN Inputs) — supportive finding docs; use when formulating product docs, rationale, or domain-facing copy 7. `SANDBOX.md` on the delivery branch — promotion input for the sandboxed element; production paths follow its Promote map 8. User paste Resolve issue: user key/URL, or ask once "Which issue should this implementation track?" ## Tracker status Follow the [implementation rows](../workflow/tracker-sync.md#matrix), including Sub-task transitions and the enabled ISSUES mirror. The parent Task remains open for ship closeout. ## Steps Follow the CONCEPT_IMPLEMENTATION flow with these specialisations: 1. **Resolve inputs** — Apply each Inputs row. Then apply the implementation start transition. A contained element that needs inspect-each-turn is not promotion-ready until `SANDBOX.md` has a complete `## Representativeness` and the last verdict is accept. Apply CONCEPT_SANDBOX in this invocation. On a delta, ask the one question and end the turn with the output unfinished (the workflow keeps **Next** on this skill). On accept, promote and continue. If the user ends sandbox-only, stop with outcome `end`. When the spec is `ADOPT.md` and the behaviour map is missing, still has `gap` rows, or a startable area has no working-surface rows, apply [../adopt/characterize.md](../adopt/characterize.md) in this invocation until the map is locked. Done when the usable spec, `ARCHITECTURE.md`, binding params, and active packages are known and the Task is **In Progress**. 2. **Resolve delivery and commands** — Follow [delivery continuity](../workflow/delivery.md) and inspect repository-owned test/lint commands. Done when the Task's one delivery head is checked out in this session's working tree and verification commands are recorded. 3. **Execute packages** — If `implement.mode=multiagent`, use CONCEPT_DELEGATION for workers; if `single`, keep Routine packages on the manager when safe. Include [testing.md](testing.md), [structure.md](structure.md), and [STRUCTURE-CATALOG.md](../concepts/STRUCTURE-CATALOG.md) in briefs. Pass-criteria rows from the definition artifact go in the brief. Fail a package that omits **spec locks** for those rows (testing.md). When spec is `ADOPT.md`, also include the Behaviour map, lock-suite commands, and working-surface commands; fail a package that rewrites lock-test expectations or that skips working-surface proof when the area owns a startable surface. Fail a package that still breaches the structure catalog, that omits the structure/testing report, or that defers catalog work to harden or lasers. Change size does not relax the briefs. When packages write product docs or domain-facing copy, pass `RESEARCH.md` / `MODEL.md` paths as brief inputs. When promoting a sandbox, follow `SANDBOX.md` Promote map into production paths. When a remaining package is a contained element that needs inspect-each-turn, apply CONCEPT_SANDBOX in this invocation until promotion-ready, then promote. When `implement.verify=comparative`, follow [rework.md](rework.md) (baseline → candidate → compare → reiterate when `implement.iteration=until-bar`). Done when all packages satisfy the spec, pass criteria, verification mode, structure catalog, Sub-task criteria, and binding. 4. **Closeout gate** — Walk the **whole** delivery diff against [structure.md](structure.md#manager-gate-before-leaving-implement) and [testing.md](testing.md) (including **Working surfaces** and **spec locks**). Remaining catalog breaches, missing reports, missing spec locks, missing tests, or missing working-surface proof → re-delegate and stay in this skill. Done when the gate holds or every remainder is a documented exception. 5. **Verify and deliver** — Run the recorded checks, update the same PR (include binding summary, test plan, and structure notes), and apply the implementation tracker row. Keep the Task **In Progress**. Record outcome `built`, or `fixed` when this invocation was fix-forward (return the Task to **In Review**). Apply the workflow transition. Done when checks pass, the gate holds, and the PR and mirrors are current. ## Work packages | Type | Subagent | Default category | Elevate when | |------|----------|------------------|--------------| | Structure exploration | `explore` | Mid | Unfamiliar large area with ambiguous seams → high | | Research | `generalPurpose` | Mid | Novel domain spike with conflicting approaches → high | | Implementation | `generalPurpose` | Mid (Routine → low) | Novel design, security/authz, concurrency, large cross-cutting → high | | Testing | `generalPurpose` | Mid (Routine → low) | Flaky tests, concurrency tests, subtle regression isolation → high | | Harden | `generalPurpose` | Mid (Routine → low) | Large structural split, layer inversion, cycle break → high | | Comparative eval | `generalPurpose` | Mid | Novel eval setup, control/performance isolation, conflicting metrics → high | | Promote | `generalPurpose` | Mid (Routine → low) | Sandbox spans many production seams or public API → high | | Fix-forward | `generalPurpose` | Low (obvious) / Mid otherwise | Architectural must-fixes, subtle correctness/races, prior lower-tier miss → next tier / high | Ensure each behavioural package lists spec locks and structure notes; if the plan omitted verification, add Testing packages before verify. For comparative verify, ensure Baseline / Compare / Reiterate packages exist per [rework.md](rework.md). A Harden package during Build repairs catalog breaches in-package; it does not replace `/restructure`. ## Inputs | Input | When present | When absent | |-------|----------------|-------------| | Definition spec (`PLAN.md`, `BUG.md`, `TWEAK.md`, `REFINE.md`, `REWORK.md`, `ITERATE.md`, or `ADOPT.md`) | Apply it, including pass criteria and any Workflow binding | Define it in this invocation (CONCEPT_ALIGNMENT, CONCEPT_DEFINITION, CONCEPT_CLASSIFICATION) and write that spec, then apply it | | `ARCHITECTURE.md` | Apply the shape. Do not ignore it | Apply CONCEPT_ARCHITECTURE in this invocation, write `ARCHITECTURE.md`, then apply it | | `RESEARCH.md`, `MODEL.md` | Apply when writing product docs or domain-facing copy | Continue without them | | `SANDBOX.md` | Apply the promote map when promotion-ready | When the element needs an inspect-loop, define the sandbox in this invocation until accept | ## Output delivery diff — packages on the delivery branch/PR, with spec locks and the structure gate held. Outcome: `built`, or `fixed` for fix-forward. This skill does not name a successor. Apply the workflow transition before the turn ends. Do not skip a closeout phase from this skill; the bound chain decides.
From marcuskrogh/skills, MIT.
More from marcuskrogh/skills
- 0.587score
help
marcuskrogh
Help choosing skills and workflows.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
iterate
marcuskrogh
Iteration on already merged work.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
adopt
marcuskrogh
Adopt the structure catalog across an existing codebase that was not built to it.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
define
marcuskrogh
Definition front door for concrete work.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
explore
marcuskrogh
Exploration through foggy or oversized work.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
refine
marcuskrogh
Refine alignment and lightweight definition for a limited area that needs refactoring or descriptive improvement without changing behaviour.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
rework
marcuskrogh
Rework alignment and lightweight definition for an intentional implementation change that must not degrade measured outcomes.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
model
marcuskrogh
Mathematical alignment through LaTeX-only questions.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
restructure
marcuskrogh
Restructure: refactoring pass against the structure catalog (campground, names, size, cohesion, seams, named smells, CRAP).
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
review
marcuskrogh
Review of a change as sequential or bundled lasers across Spec, Correctness, Integration, Architecture, and Standards: find, fix, then publish a code review.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
architect
marcuskrogh
Architecture for one Task: record where the work sits (modules, layers, seams, dependency direction) in ARCHITECTURE.md.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
bug
marcuskrogh
Bug alignment and lightweight definition for a clear defect.
SkillAFreshness: A★ 1Claude Code +9
Signals
- Stars
- 1
- Forks
- 1
- Open issues
- 0
- Last commit (this folder)
- Sep 27, 2026 (1 month ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/marcuskrogh/skills
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from marcuskrogh
- 0.587score
iterate
marcuskrogh
Iteration on already merged work.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
adopt
marcuskrogh
Adopt the structure catalog across an existing codebase that was not built to it.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
define
marcuskrogh
Definition front door for concrete work.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
explore
marcuskrogh
Exploration through foggy or oversized work.
SkillAFreshness: A★ 1Claude Code +9 - 0.587score
help
marcuskrogh
Help choosing skills and workflows.
SkillAFreshness: A★ 1Claude Code +9 - 0.584score
rework
marcuskrogh
Rework alignment and lightweight definition for an intentional implementation change that must not degrade measured outcomes.
SkillAFreshness: A★ 1Claude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get implement running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
implement skill. Agent Mart. https://agentmart.dev/skills/marcuskrogh-implement (last updated 2026-10-24)