| name | twg |
| description | Root TWG CLI skill for Atlassian work-data tasks. Use typed commands for known anchors; use live `twg help` only when command shape or output contract is uncertain.
|
twg
First TWG routing step: use typed commands for stable keys, URLs, ARIs, and
familiar families. If uncertain, inspect twg help <terms> or
twg help describe <path>. For specialized guidance, run
twg help discover-skills "<intent>".
Overview
For synthesis, load the narrowest workflow skill:
twg-status-rollups for person, team, org, project, goal, leadership, and appraisal readouts.
twg-context-discovery for topic deep dives, dependency maps, context graphs, repo discovery, and catch-ups.
twg-engineering-work for PR queues, stale reviews, repo contributors, hot
areas, and PR status.
twg-jira-resolve-merged-work for dry-run-first Jira board/sprint/epic/space cleanup where merged PR evidence can safely resolve stale workitems.
twg-operational-health for handoffs, reliability/incidents, Assets,
staffing, meeting summaries, and risk.
twg-bench-lite for read-only single-prompt A/B comparisons.
Invocation And Output
Run TWG directly:
twg <command>
Agent hosts may set TWG_AGENT_DEFAULTS=1; do not add per-command env prefixes unless requested.
For large outputs, inspect output_files.compact first when present. Read or
filter output_files.stdout only when compact output lacks evidence.
Auth/Setup Guard
Do not run setup, login, logout, install, update, upkeep, or credential commands
from ordinary TWG requests. Only run them for explicit setup/auth/repair
asks or setup/auth skills. If missing, report remediation and wait for user direction.
Bounded Evidence Loop
For synthesized answers, converge instead of exhaustively hunting; prefer typed
commands or product-native paths with stronger evidence.
- Classify the anchor: person, team, project, goal, workitem, page, repo,
service, asset, or topic.
- Resolve IDs once, then fetch only evidence that can change owner, status,
risk, decision, relationship, priority, or next action.
- Group large candidate sets; hydrate representative top items.
- Prefer compact/evidence output. Inspect full stdout only for missing fields.
- After each batch, synthesize once evidence, recency, and confidence hold.
- Stop a path after the same auth, ACL, contract, or backend error twice.
Command Discovery
- Use typed commands directly for familiar families:
resolve, search,
user, org-tree, work query, work search, pull-requests, jira,
confluence, docs, context, responsibility, goals, projects,
assets, and trello.
- Use
twg search "<topic>" [--limit <n>] for relevance-ranked top-K discovery
across built-ins and ready connectors. Implicit unavailable connectors warn;
explicit --app preflights and fails.
- For fuzzy Trello board/card discovery, use
twg trello search "<query>" --limit 20; no workspace scoping.
- For Rovo connector searches, use
twg rovo list-apps -o json
(list-connectors is an alias) to discover site-visible --app filters.
Explicit twg rovo search ... --app <connector> preflights connector auth;
follow the returned action or run twg rovo auth <app>, then retry.
- Keep document relationship history and fuzzy discovery separate:
- Use
twg docs query --since <duration> [--account-id <id>] [--first <n>]
for documents related to a user through activity, not title/content search.
- Use
twg docs search "<topic>" [--limit <n>] for fuzzy Rovo search across
Confluence and ready document connectors.
- Never pass topic text to
docs query; route that intent to docs search.
- Keep user activity and fuzzy work discovery separate:
twg work query defaults to seven days of authored work, full-window counts, and five summary items.
- Other activity is opt-in through
--activity / --include-viewed.
twg work search "<topic>" is tenant-wide work discovery; use docs search for documents.
- If fuzzy topic text reaches
work query, the CLI redirects to work search; prefer direct calls.
- Use
twg help <terms> before guessing grammar and twg help describe <path> for contracts.
- Namespace help is not an executable contract. Follow advertised child commands
before adding flags.
- Resolve URLs, keys, ARIs, names, and people first; hydrate stable IDs with
product-native commands.
- For Jira discovery, keep the three search modes distinct: use
jira workitem search <text...> for Jira-only fuzzy text backed by JQL,
jira workitem query --jql <jql> for explicit structured filters, and
search <text...> --app jira for semantic Rovo discovery.
- Command shape guardrails:
- Known Jira/Atlas keys are positional for
jira workitem get, goals get,
and projects get; query --key forms are compatibility shortcuts.
work query is user activity only (--scope me|user); use work search
for topic discovery and never use --scope global.
work search "<topic>" is fuzzy work discovery. Use advertised filters such as --types.
assets search is shallow. For device/person joins, inspect schemas/types,
shortlist owners, then batch assets query/assets object query with repeated --account-id.
- Keep projection commands and product-native commands separate. Do not borrow
flags across surfaces unless live help advertises them.
- Use
search-code for indexed code. --app selects bbc, github, or
gitlab; --workspace maps to workspace/org/group.
Load The Narrowest Companion
Use a concrete key, URL, ARI, slug, account ID, name, topic, me, or window.
Load companion skills for detailed semantics:
../twg-jira/SKILL.md for Jira semantics.
../twg-confluence/SKILL.md for Confluence semantics and safe body editing.
../twg-status-rollups/SKILL.md for personal/team/org/project/goal/leadership status.
../twg-context-discovery/SKILL.md for context, dependencies, responsibility, and graphs.
../twg-engineering-work/SKILL.md for PRs, reviews, contributors, and bottlenecks.
../twg-jira-resolve-merged-work/SKILL.md for board/sprint/epic/space stale Jira workitems whose implementation appears complete from merged PR evidence.
../twg-operational-health/SKILL.md for handoffs, reliability, incidents, assets, staffing, and risk.
Rules
- Never guess IDs, flags, account IDs, page IDs, repo slugs, ARIs, object IDs,
or mutation contracts.
- For product-specific writes or rich body formats, load the relevant product
skill first and follow exact live help.
- Do not route normal TWG prompts through local workspace inspection,
cached-output spelunking, raw schema probing, or process diagnostics unless the
user asks about local state.
- For writes, read current state first and state the intended mutation unless
the user explicitly asked you to execute it.