| name | org-work-sensing |
| description | Assess what is happening inside an organization or team by inspecting systems of work such as GitHub, Jira, Linear, issue trackers, pull requests, project boards, incidents, discussions, and operating docs. Use when a user asks for a pre-work assessment before adopting agentic workflows, current-state and gap analysis, org health readout, delivery/process diagnosis, bottleneck analysis, team operating assessment, cross-functional work review, or recommendations based on evidence from messy or incomplete work artifacts.
|
Org Work Sensing
This skill helps Codex inspect work artifacts, infer operating patterns, and
recommend practical changes. Treat it as an assessment layer: it diagnoses what
the work signals suggest, identifies what must be fixed before automation is
useful, then routes setup or automation recommendations to other skills when
needed.
Core Workflow
- Clarify the assessment frame when it is not obvious:
- organization, team, repo, project, or timeframe
- systems of record available, such as GitHub, Jira, Linear, Notion,
Confluence, Slack exports, incident tools, or project boards
- access path for each system: MCP/connector, CLI, API, local repo,
exported CSV/JSON, copied text, or representative links
- desired output: executive readout, team retro, workflow recommendations,
readiness gap analysis, risk review, automation candidates, or a
prioritized action plan
- Inventory available access before analysis. Do not assume MCPs or connectors
are configured. If direct access is missing, request exports or representative
artifacts and record the access gap as part of the current-state assessment.
- Collect evidence from accessible sources. Prefer direct artifacts over
recollection:
- GitHub issues, PRs, discussions, milestones, labels, projects, releases,
workflow runs, and CODEOWNERS or policy files
- Jira or Linear issues, statuses, cycle metadata, comments, components,
labels, epics, and blockers when available
- operating docs such as strategy, roadmap, how-we-work, incident reviews,
decision logs, support escalations, launch plans, and compliance notes
- Read
references/assessment-signals.md when doing a substantive assessment
or when comparing evidence across tools.
- Separate observations, inferences, and recommendations:
- observation: directly supported by artifacts
- inference: likely pattern or cause, with confidence and caveats
- recommendation: concrete change, owner, expected effect, and first step
- Produce an assessment that is useful without overclaiming. Include missing
evidence, contradictory signals, and any access limitations.
- Identify readiness gaps before recommending agentic workflows:
- missing systems of record
- unclear work taxonomy or ownership
- sparse or unreliable status data
- absent strategy, how-we-work, decision, or intake artifacts
- tool topology that prevents a useful agent from seeing the work
- If the recommendation is to create workflows, docs, issue templates, policies,
project boards, or repo setup, route implementation through
non-coder-agentic-workflow-builder. If the recommendation requires GitHub
Agentic Workflow creation or debugging, route through agentic-workflows.
Evidence Standards
- Prefer bounded time windows. Default to the last 30-90 days unless the user
asks for a longer trend.
- Do not infer individual performance from sparse artifacts. Focus on system
design, work flow, decision quality, ownership, and coordination patterns.
- Treat private, sensitive, or HR-adjacent information carefully. Summarize
patterns instead of exposing unnecessary personal details.
- State data source coverage. Name which repos, projects, issue queries, docs,
or exported files were inspected.
- Treat missing connector access as a finding, not a failure. Explain what can
and cannot be assessed from the available artifacts.
- Use counts and examples when available, but do not let metrics replace
qualitative evidence.
- Preserve uncertainty. Use confidence labels such as high, medium, or low when
the evidence base is partial.
Output Pattern
For most assessments, use this structure:
- Scope and sources inspected
- Executive readout
- Current-state topology
- Key operating signals
- Readiness gaps
- Bottlenecks and risks
- Recommendations
- Suggested workflows or artifacts to add
- Evidence gaps and next questions
Keep recommendations practical. Prefer small operating changes before broad
reorganizations or heavy automation.
For pre-work assessments, include a readiness ladder:
- Minimum viable visibility: what an agent or human can reliably inspect today
- Missing substrate: docs, labels, fields, issue templates, ownership, or links
- Cleanup sequence: the smallest fixes that make the system legible
- Automation candidates: only the workflows that are ready enough to implement
Recommendation Heuristics
- If work is stale or blocked, check ownership, decision paths, review queues,
dependencies, and unclear priority before recommending more process.
- If many issues are created but few close, inspect intake quality, triage
rituals, label hygiene, capacity mismatch, and whether work items map to
actual decisions.
- If PRs wait a long time, inspect review ownership, CODEOWNERS, CI stability,
batch size, review norms, and release pressure.
- If priorities drift, inspect strategy docs, roadmap links, project fields,
milestone discipline, and repeated escalations.
- If stakeholders ask the same questions repeatedly, recommend living status
reports, decision logs, launch readiness reviews, or leadership briefs.
- If the team lacks usable source material, recommend starting with strategy and
how-we-work docs before adding automation.
- If the topology is messy, map the current systems first: where work is
requested, prioritized, executed, reviewed, released, and remembered.
- If the system is not ready for agentic workflows, say so directly and propose
pre-work rather than forcing an automation design.
Tool Notes
- For GitHub, use
gh and local repo inspection when available. Avoid write
operations unless the user explicitly asks for implementation.
- For Jira, Linear, or other trackers, first look for available MCPs,
connectors, CLIs, exports, or user-provided files. If direct access is
unavailable, ask for CSV/JSON exports or representative issue links.
- For generated recommendations that become repo changes, make those changes in
the target repo and validate with the appropriate repo checks.