agentsclimarketplace

Abbie project work

Skill creative-int/abbie-plugins/skills/abbie-project-work

Portable Abbie skills and credential-safe MCP wiring for agent clients.

Install
npx -y skills add creative-int/abbie-plugins --skill abbie-project-work

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 13 days oldThe repository was created 13 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

This skill should be used when setting up or inspecting an Abbie Project, binding its stable Workspace source truth, starting or continuing a governed Task, or following work through Result, Changes, Checks, and Proof. Triggers include 'set up an Abbie project', 'bind this repo', 'run this as an Abbie task', 'check task progress', and 'continue this task'.

SKILL.md

5.2 KB, ~1.2k tokens by cl100k_base, as published. Nobody here has run it

Abbie Project Work

Use Abbie as the authority for project identity, Workspace source truth, and durable Task outcomes. Keep the sequence intact: Project → Workspace → Task → Result · Changes · Checks · Proof.

Preconditions

  1. Confirm the product CLI is connected:

    abbie account status --json
    abbie mcp status --probe --json
    
  2. If the account is disconnected, ask the user to run:

    abbie account login
    
  3. Never request, copy, print, or place an Abbie token in a plugin configuration. The local abbie mcp serve bridge owns credentials.

Choose the Surface

  • Prefer the abbie_projects, abbie_workspace, and abbie_tasks MCP tools when they are mounted in the current agent.
  • Use the public abbie CLI when shell access is available or when the task requires local source inspection or local application.
  • Use --json for machine-readable CLI calls.
  • Do not invent a second project, workspace, task, or proof schema.

Workflow

1. Resolve the Project

List Projects before creating one:

abbie project list --json

With MCP, call abbie_projects with action: "list".

Reuse a Project when its identity and repository match. Create only when no matching Project exists:

abbie project create \
  --name "My project" \
  --repo owner/repository \
  --branch main \
  --profile workspace-write \
  --json

With MCP, call abbie_projects with action: "create", name, repo_url, repo_branch, and default_task_profile.

Capture the returned Project ID. Treat it as the stable identifier for every following step.

2. Establish Workspace Source Truth

Inspect the Project's one stable Workspace:

abbie workspace show --project <project-id> --json

Bind it only when the intended repository, branch, or directory is absent or wrong:

abbie workspace bind \
  --project <project-id> \
  --repo owner/repository \
  --branch main \
  --directory /absolute/repository/path \
  --json

From the bound repository, report current local source truth:

abbie workspace status --project <project-id> --json

The status command checks repository and branch identity before reporting the revision, dirty state, ahead/behind counts, and sync state. Do not start workspace-writing work while the binding and local checkout disagree.

MCP clients may use abbie_workspace with show, bind, or report. Only a local CLI can inspect the current Git checkout automatically.

3. Start the Governed Task

Use a prompt with a concrete outcome and proof boundary:

abbie task start \
  --project <project-id> \
  --profile workspace-write \
  --prompt "Implement the requested change and report Result, Changes, Checks, and Proof." \
  --json

Use safe for read-only or diagnostic work and workspace-write only when changes are intended. With MCP, call abbie_tasks with action: "start", project_id, prompt, and safety_profile.

Capture the returned Task ID.

4. Follow the Task to a Decision Boundary

For a blocking CLI wait:

abbie task wait <task-id> --json

For MCP, poll abbie_tasks with action: "get" and task_id. Avoid rapid polling; use a short interval and stop at a handoff or terminal state.

When the Task requests clarification, surface the question without guessing. When the Task needs a changed brief, continue the same Task:

abbie task continue <task-id> \
  --prompt "Refined instruction" \
  --json

Do not create a replacement Task merely to answer a follow-up.

5. Read the Four-Part Outcome

Inspect the Task:

abbie task get <task-id> --json

Report each part distinctly:

  • Result — what the Task accomplished.
  • Changes — the exact authoritative change set.
  • Checks — verification that ran and its outcome.
  • Proof — durable receipts for the result and actions.

If a verified change set exists, inspect it separately:

abbie task changes <task-id> --json

Do not describe work as applied or published merely because a change set was prepared. Use abbie-proof-review for the mutation decision.

Completion Contract

Return:

  • Project name and ID
  • Workspace binding and reported source state
  • Task ID and status
  • Result
  • Change-set status and fingerprint, if present
  • Checks
  • Proof receipt identifiers
  • The next explicit decision, if any

Guardrails

  • Never bypass a required confirmation.
  • Never claim local files changed from an MCP apply preparation alone.
  • Never infer a draft pull request from a prepared receipt; require its URL or explicit published state.
  • Never expose credentials or local secret storage.
  • Never collapse unknown, pending, stalled, or needs-review into success.

Reference

Read references/project-work-contract.md for command and MCP mappings, state semantics, and handoff examples.

What ships with it: 1 file

3.7 KB alongside SKILL.md

references/

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.