
Twg Context Discovery
- 7 installs
- 8 repo stars
- Updated August 5, 2026
- atlassian/twg-cli
twg-context-discovery is a TWG CLI workflow skill that builds context graphs, dependency maps, and topic deep dives from Atlassian work data.
About
This TWG CLI workflow skill produces topic deep dives, dependency maps, and context graphs over Atlassian work data. A developer uses it to answer 'catch me up', 'which repo should I change', or 'who is involved' by hydrating a central anchor and its relationship edges before synthesizing. It expands by relationship role and can pipe typed context output to twg visualize to render a focused graph.
- Builds topic deep dives, dependency maps, and context graphs from Atlassian work data
- Resolves an anchor first, then expands by relationship role, not raw count
- Pipes typed context output to twg visualize to render focused graphs
Twg Context Discovery by the numbers
- 7 all-time installs (skills.sh)
- Ranked #2,276 of 3,282 Productivity & Planning skills by installs in the Skillselion catalog
- Data as of Aug 5, 2026 (Skillselion catalog sync)
twg-context-discovery capabilities & compatibility
- Capabilities
- context discovery · dependency mapping · work data query
- Works with
- atlassian · jira · confluence · bitbucket
- Use cases
- research · project management
What twg-context-discovery says it does
Use with the root `twg` skill for topic deep dives, dependency maps, graph
Expand by relationship role, not raw count.
npx skills add https://github.com/atlassian/twg-cli --skill twg-context-discoveryAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Installs | 7 |
|---|---|
| repo stars | ★ 8 |
| Last updated | August 5, 2026 |
| Repository | atlassian/twg-cli ↗ |
What it does
Build topic deep dives, dependency maps, and context graphs over Atlassian work data with the TWG CLI.
Who is it for?
developers who need the full context and dependency map around a work item or topic
Skip if: raw graph-query debugging, which the skill says is not the default dependency-map route
When should I use this skill?
the user asks to catch up on a topic, map dependencies, find the right repo, or see who is involved
What you get
A hydrated context graph with anchor snapshot, relationship table, risks, and next actions for a topic or work item.
- anchor snapshot
- relationship table
- risk and dependency list
By the numbers
- escalates full detail to at most 3 high-signal related anchors
Files
twg-context-discovery
Use together with the root twg skill. Exact command grammar must come from live twg help, twg help <terms>, or twg help describe <path>.
Use When
- "Catch me up on..."
- "Deep context around this workitem/page/topic"
- "Dependency map" or "find upstream/downstream blockers"
- "Which repo should I change?"
- "Which people/repos/projects are related?"
- "Draw the graph" / "visualize" / "open the graph"
- "Who's involved" / "experts on X"
First Move
Resolve the anchor before widening:
- Stable key, URL, or ARI: use it directly when the command family is clear.
- Fuzzy topic or name: use resolve/search once, then select concrete anchors.
- Multiple same-kind anchors: batch them in one context call when supported so
edges return together.
- Unknown command shape: inspect one focused help contract before calling data.
If a context command is not advertised for an anchor type, treat that as a coverage gap. Use product-native hydration and search evidence instead of inventing unsupported command paths.
Route Selection
- Known Jira work items usually need native workitem details plus relationship
context.
- Projects and goals usually need native project/goal details plus selected
Jira, Confluence/docs, search, PR, and meeting evidence.
- Pages/topics/dependency prompts need hydrated central anchors before broad
search results become evidence.
- Raw graph-query/debugging surfaces are not the default dependency-map route.
Use them only when the user explicitly asks for that query language or typed commands cannot express the required edge.
Evidence Policy
For every central candidate, run source fetch and relationship context as one bounded fan-out:
- Source fetch: fields, owner, status, body, comments, and URLs.
- Context: graph edges, formal external links, related people, teams, projects,
goals, docs, PRs, commits, and branches.
Use summary detail first. Escalate to full only for the central anchor or up to 3 high-signal related anchors when URLs, comments, body content, or provenance are missing.
Read output_files.stdout before selecting entities for graph/context answers; stdout_shape is only a structural sample and can miss relationship tails.
Third-party URLs are first-class graph nodes. Collect URLs from remote links, formal context external edges, descriptions, comments, ADF link marks, bare URL text, and linked target bodies. Track provenance because it determines relationship direction.
Expansion Rules
- Expand by relationship role, not raw count.
- Hydrate parent, epic, inbound peer, blocker, consumer, central page, external
design, PR, commit, branch, assignee, reporter, contributor, and reviewer signals when they change direction, risk, ownership, or next action.
- Fetch known older links directly by URL, key, ID, or ARI instead of widening
the whole graph blindly.
- Use strong query variants rather than many synonyms.
- Stop when the next candidate would not add new entities, links, contributors,
teams, decisions, ownership, risk, or next action.
Graph Visualization
When the user asks to visualize, draw, or open a graph, prefer piping typed context output to twg visualize. The viewer can auto-detect context envelopes and project them to graph shape. Hand-author graph JSON only when typed projection cannot express the needed labels or merged multi-anchor graph.
Keep rendered graphs focused on directly relevant people, teams, and artifacts. Collapse duplicate signals and keep peripheral items out of the graph when they do not change direction, ownership, risk, or next action.
Output Shape
- Anchor snapshot: what it is and why it matters.
- Relationship table: entity, type, direction, owner, importance, and evidence.
- Risks and dependencies, separating confirmed edges from inferred relationships.
- Suggested next actions.
- Confidence and gaps when evidence is incomplete, access-limited, stale, or
sampled.
Anti-Patterns
- Do not stop at search results without hydrating anchors.
- Do not treat
stdout_shapeas a complete entity or URL inventory. - Do not skip peer expansion for graph/dependency prompts because peers look
"Done".
- Do not dismiss a 1-hop candidate by title alone.
- Do not hand-roll graph HTML.
Related skills
FAQ
What is the first move in context discovery?
Resolve the anchor before widening: use a stable key/URL/ARI directly, or resolve/search once for fuzzy topics.
How does it decide what to expand?
It expands by relationship role, not raw count, and stops when the next candidate adds no new entities, risk, or next action.