fullsend

name

fullsend

description

Operates Fullsend from a repository where it is already installed. Discovers the repository's actual Fullsend installation, available agents, harnesses, skills, workflows, and configuration sources before helping users trigger, watch, or inspect agent runs. Use when asked what Fullsend can do in the current repo, which agents or skills are available, how an agent is configured, to run an `/fs-*` agent, monitor a run, investigate a failed run, or find the relevant upstream Fullsend user guide. This is not an installation, administration, upgrade, or agent-authoring skill.

Fullsend

Use the Fullsend installation present in the current repository. Treat local configuration and pinned workflow references as evidence; do not assume that a stock agent set or a particular deployment layout is installed.

<essential_principles>

Essential principles

1. Orient before acting. Load the repository's Fullsend context before every task unless it is already current in this conversation. Agent names, skills, workflow paths, and configuration sources vary by installation.
2. The checkout is the starting point. Resolve the git root from the current directory, then inspect its .fullsend/ content and Fullsend workflow files. Follow pinned reusable-workflow or agent sources only when local files do not answer the user's question.
3. Report observed facts separately from inference. Label an agent or skill as available only when the installation exposes it. If a remote source cannot be read, report the source and the resulting uncertainty.
4. Stay user-facing. Help developers understand and operate an existing installation. Do not install or uninstall Fullsend, edit infrastructure, upgrade the CLI or scaffold, manage org enrollment, or build/customize agents. Point those requests to the canonical upstream guides.
5. Confirm shared-state writes. Show the exact repository, issue or PR, and comment body before posting a slash command. Read-only inspection needs no confirmation.
6. Prefer canonical documentation. Link to Fullsend's upstream guides rather than reproducing setup, configuration, or workflow documentation here.

</essential_principles>

Mandatory orientation

Run:

python3 scripts/orient.py --repo <checkout-path>

Consume the complete JSON output. Do not filter it through head, tail, grep, or jq. If the script cannot find Fullsend markers, state that this checkout does not appear to use Fullsend and ask for the correct repository.

Read references/orient.md when the user asks about available agents, skills, configuration, or when the local report contains unresolved remote sources.

Commands

Command	Purpose
orient [repo-path]	Explain the installation, agents, skills, and configuration sources
trigger <agent> <issue-or-PR>	Post a verified /fs-* command
watch <issue-or-PR-or-run-id>	Monitor a run until it finishes
inspect [issue-or-PR-or-run-id]	Summarize run status, output, and failure evidence
help [topic]	Link to the relevant canonical user documentation

If the user already expressed an intent, route directly. If invoked without an intent, show this table and ask what they want to do.

Routing

Intent	Reference
orient, agents, skills, config, how is this installed	references/orient.md
trigger, run, /fs-*	references/trigger.md
watch, wait, monitor	references/watch.md
inspect, status, logs, failed run, what happened	references/inspect.md
help, docs, guide, run locally	references/help.md

Load only the routed reference. Orientation is shared context, not a separate user-visible prerequisite command.

<repo_resolution>

Repository resolution

Use this order:

1. An explicit path or --repo owner/name supplied by the user.
2. The git root containing the current working directory.
3. A parent directory containing .fullsend/ or a Fullsend workflow.

For GitHub operations, derive owner/name from git remote get-url origin or gh repo view --json nameWithOwner. Never use organization-specific fallback paths or repositories.

</repo_resolution>

Out-of-scope requests

For installation, administration, configuration changes, agent authoring, or upgrades, do not improvise a procedure. Explain that this skill is scoped to users of an existing installation and link the matching page from the upstream guide index.

Session artifact importing and transcript visualization remain separate from this skill.