| name | ccdash |
| description | Drive the standalone CCDash CLI from natural language to answer questions about project status, feature forensics, agent sessions, workflow failures, and after-action reports. Use when the operator or another agent asks about CCDash project health, agent-session analytics ("why did FEAT-X take so long", "what sessions ran for this feature", "which workflows are failing"), wants an AAR / narrative feature report, or needs help installing / authenticating the `ccdash` CLI against a target server. Routes intent to `ccdash` subcommands with the right `--output` mode, echoes provenance IDs for chaining, and degrades gracefully through `ccdash doctor` on transport/auth failures. Do not use for unrelated coding tasks, writing UI components, or anything outside CCDash project/session intelligence. |
ccdash Skill
A thin natural-language router over the standalone ccdash CLI. The CLI is authoritative; this skill decides which subcommand to run, which --output mode to pick, and how to chain calls using provenance IDs.
When To Use
Trigger on intents such as:
- "What's the state of this project?" / "project status" / "project summary"
- "Why did FEAT-X take so long?" / "retrospective on FEAT-X" / "AAR for FEAT-X"
- "Which workflows are failing?" / "failure burden" / "flaky workflows"
- "Show sessions for FEAT-X" / "search session transcripts for Y" / "drilldown sentiment on session Z"
- "Install ccdash" / "set up ccdash against staging" / "log in to the staging target"
- Connectivity / auth error messages mentioning
ccdash, CCDASH_TARGET, or an HTTP failure against a CCDash server.
When NOT To Use
- General coding tasks unrelated to CCDash observability (building UI, editing backend code, unrelated debugging).
- Questions about other CLIs or platforms even if phrased as "project status" (e.g. Linear, GitHub, Vercel).
- Authoring new CCDash reports, workflows, or analytics queries — this skill only invokes existing commands.
- MCP tool invocation — deferred until the CCDash MCP surface ships; see the parent enablement plan (Phase 4 of this skill).
Preflight (run once per session on uncertainty)
ccdash --version
ccdash target show
If ccdash is not on PATH or target show errors: follow recipes/target-onboarding.md.
If a subsequent command returns a connection or 401/403 error: follow recipes/unreachable-server.md before surfacing the raw error.
Routing
The authoritative intent -> command map lives in scripts/router-table.json. Load it for disambiguation; the summary below is a lossy render kept for quick scanning.
| Intent | Command | Default --output | Reference |
|---|
| Project status | ccdash status project | json (agent) / human (operator) | references/command-status.md |
| Doctor / diagnose | ccdash doctor | human | references/command-doctor.md |
| Install / onboard | pipx install ccdash-cli, then target add | human | references/install-setup.md |
| Target management | ccdash target {list,add,use,show,remove,login,logout,set-token,check} | human | references/command-target.md |
| Workflow failures | ccdash workflow failures | json | references/command-workflow.md |
| Feature list / show / sessions / documents | ccdash feature {list,show,sessions,documents} | json | references/command-feature.md |
| Session list / show / search / drilldown / family | ccdash session {list,show,search,drilldown,family} | json | references/command-session.md |
| AAR / narrative report | ccdash report {aar,feature} | markdown | references/command-report.md |
Output-mode decision rules live in references/output-modes.md. Provenance fields to echo back into agent context live in references/provenance.md.
Multi-Step Flows (Recipes)
Multi-step investigations use deterministic recipes in recipes/:
recipes/project-triage.md — status project -> risky feature -> feature show -> feature sessions.
recipes/feature-retrospective.md — feature show -> feature sessions -> report aar.
recipes/workflow-failure-rootcause.md — workflow failures -> pick workflow -> session drilldown --concern.
recipes/session-cluster-investigation.md — session show -> session family -> drilldown each sibling.
recipes/unreachable-server.md — transport/auth failure interpretation via ccdash doctor.
recipes/target-onboarding.md — fresh operator: install -> target add -> target login -> doctor.
Prefer a recipe whenever the user's intent requires more than a single subcommand call. Single-command intents route directly via the table above.
Output Mode Quick Rule
- Commands consumed by the agent for reasoning:
--output json (or the --json shortcut).
- Commands rendered verbatim to the user as narrative:
--output markdown (or --md). Defaults for report aar and report feature.
- Commands an operator runs in a terminal (target management, doctor): default
human.
Full rules: references/output-modes.md.
Provenance (always echo)
For every command result used in reasoning, surface the stable IDs and timestamps listed in references/provenance.md into the agent's working context so follow-up calls can chain without re-fetching.
Graceful Degradation
Any transport/auth error (connection refused, DNS, TLS, 401/403, 5xx) MUST route through ccdash doctor before being surfaced as a raw error. See recipes/unreachable-server.md.
Extending The Skill
When a new ccdash subcommand ships, do not edit this file. Follow the update protocol in the implementation plan (Update Protocol section):
- Add/extend
references/command-<group>.md.
- Append a row to
scripts/router-table.json.
- Update
CHANGELOG.md.
Edit SKILL.md only if trigger scope or when-not-to-use guidance changes.
Key Files
scripts/router-table.json — canonical intent -> command map (load on ambiguity).
scripts/preflight.sh — optional install/connectivity probe.
references/cli-overview.md — full command tree snapshot + global flags + target/auth resolution.
references/install-setup.md — pipx / pip / repo-local install + verification.
references/output-modes.md — human vs json vs markdown decision rules.
references/provenance.md — IDs, timestamps, freshness fields to echo into agent context.
references/eval-scenarios.md — 20-scenario routing eval fixture (Phase 2).
CHANGELOG.md — dated log of skill changes tied to CLI surface.