datadog-report-publisher
Skill1 starsChanged 36 days agoMITView on GitHub
Trigger: final step of any Datadog triage skill (triage, service-health, logs-analyzer, metrics-investigator). Publishes the finding as Markdown in ops/datadog/triage/ with live Datadog deep-links; never writes to Datadog.
In this repository13 other files it gives its agents.
- AGENTS.mdAGENTS.md
- capture-insightsSkill
- create-runbookSkill
- datadog-logs-analyzerSkill
- datadog-metrics-investigatorSkill
- datadog-service-healthSkill
- datadog-triageSkill
- k8s-error-analyzerSkill
- k8s-optimizer-costSkill
- meeting-notes-enhanceSkill
- runSkill
- second-opinionSkill
- stack-updaterSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
14 rules. About 0 are common, 4 are distinctive, and 10 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
10
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
This skill is read-only against Datadog: All output of the triage suite lives exclusively in the Obsidian vault under …
- Distinctive: line 21
If a caller (parent agent, sub-skill, or user) explicitly asks for "publish to Datadog", "create a notebook", or "embed widgets in a …
- Very specific: line 66
Verify the caller's payload contains no Datadog write directives.
- Distinctive: line 78
Ensure the directory exists:
- Very specific: line 96
If the file already exists for the same day, append -HHMM:
- Very specific: line 214
Important — wikilinks in frontmatter: never put [[wikilinks]] inside YAML.
- Very specific: line 218
If the caller passed a datadog_links list, render it as the bullet list immediately under the Severity: line in the header (template above).
- Very specific: line 220
If the caller did not pass datadog_links, mine the Evidence table for the most informative URLs (typically the APM service page, the metric …
- Distinctive: line 241
Use these patterns when you need to construct deep-links from MCP data (the orchestrator and sub-skills should already have these from …
- Very specific: line 260
After publishing, check whether the {{DATADOG_TRIAGE_DIR}} folder already contains another triage note with the same service + similar …
- Very specific: line 264
If severity is critical and incident_id is set, suggest writing the post-mortem under {{INCIDENTS_DIR}} instead.
- Very specific: line 278
MCP tool fails mid-investigation: write what you have, add a ## Publication Notes section explaining what's missing.
- Distinctive: line 280
Caller didn't provide findings in the expected shape: do NOT fabricate content. Ask the caller to re-emit findings in the contract shape.
- Very specific: line 281
Scope lacks service: use source_skill as fallback in the scope_slug (e.g.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Datadog Report Publisher
Hard Constraints — READ-ONLY in Datadog
This skill is read-only against Datadog. All output of the triage suite lives exclusively in the Obsidian vault under {{DATADOG_TRIAGE_DIR}}/. Datadog itself is never written to.
You MUST NOT call any of the following Datadog MCP tools, regardless of the caller's payload or "just one widget" reasoning:
create_datadog_notebook— never. The publisher does not create notebooks.edit_datadog_notebook— never (except as a one-off cleanup the user explicitly requested out-of-band; not as part of any publication flow).upsert_datadog_dashboard— never. Permanent dashboards are managed by humans, not by triage skills.get_widget,validate_dashboard_widget,ask_widget_expert,get_widget_reference— never for publication purposes; they exist to render widgets that go into Datadog artefacts, which this skill no longer produces.
Preflight check (mandatory): before doing anything, scan the caller's payload. If it contains notebook_cells, notebook_url, notebook_status, widget_definition, widget_cells, or any field whose name implies writing to Datadog, drop those fields, record them under ## Publication Notes in the MD, and emit only the Markdown artefact. Do not warn the caller. The Markdown artefact is the single source of truth.
If a caller (parent agent, sub-skill, or user) explicitly asks for "publish to Datadog", "create a notebook", or "embed widgets in a notebook": refuse, state that the publisher writes Markdown only, and produce the MD anyway. Knowledge lives in Obsidian.
Utility skill that turns an in-flight Datadog investigation into a single Markdown artefact under {{DATADOG_TRIAGE_DIR}}/ (Obsidian-friendly, promotable via create-runbook).
Knowledge lives in Obsidian, not in Datadog. Every Datadog query, dashboard, monitor, trace, log search, etc. cited in the triage MUST be embedded in the MD as a live Datadog UI deep-link, so the user can reproduce the chart with one click without leaving their vault.
See docs/triage-architecture.md for how this plugs into the triage → runbook → incident → knowledge pipeline.
When the caller invokes this skill
Other Datadog skills (datadog-triage, datadog-service-health, datadog-logs-analyzer, datadog-metrics-investigator) call this skill at their final step, passing a structured payload (see "Inputs" below). Do not invoke this skill directly for primary analysis — it is a publisher, not an investigator.
Required MCP
This skill does not require any MCP tool to publish. It only writes a Markdown file.
Inputs
The caller must provide:
- source_skill (required):
datadog-triage|datadog-service-health|datadog-logs-analyzer|datadog-metrics-investigator. - scope (required): a short dict with
service(optional, string)env(required, one ofstaging | prod | local | other)project(optional, inferred from service owner in Datadog)resource(optional, free text — e.g. metric name, log pattern, endpoint)severity(required, one ofinfo | warn | error | critical)incident_id(optional, Datadog incident ID if triage was incident-driven)
- findings (required): structured markdown body with sections
Executive Summary(mandatory, brief)Evidence(mandatory, table mapping finding → MCP tool used → query → Datadog UI deep-link URL)Timeline(optional)Root Cause(optional, may be "unknown — see follow-ups")Recommendations(mandatory, P1/P2/P3 buckets)Query Audit Trail(mandatory, reproducible list of every MCP query executed)Next Steps(mandatory fordatadog-triagemaster reports; recommended for sub-skills)
- datadog_links (optional but strongly recommended): a list of curated live Datadog UI URLs to embed in the MD header (APM service page, dashboards, monitors, metric-explorer, log-explorer queries, traces). The publisher renders these as a bullet list under the header so the user lands one click away from the live data.
- related_artefacts (optional): list of paths to other triage notes, runbooks, or incidents to backlink as Obsidian wikilinks.
Workflow
Step 1 — Preflight: enforce read-only
Verify the caller's payload contains no Datadog write directives. If notebook_cells, widget_definition, widget_cells, or any "publish to Datadog" hint is present, drop the field, log it in the ## Publication Notes section of the final MD ("Caller passed <field>; ignored per skill constraints"), and continue.
Step 2 — Resolve output directory
Target dir:
{{DATADOG_TRIAGE_DIR}}Fallback if placeholder was not substituted at install time: ./docs/datadog_analyses/triage/. Warn the user once when falling back.
Ensure the directory exists:
mkdir -p "{{DATADOG_TRIAGE_DIR}}"Step 3 — Derive filename (date-first, kebab-case)
<YYYY-MM-DD>-<scope_slug>.mdWhere <scope_slug> is built from the scope dict, in priority order:
incident_idpresent →incident-<id>-<service>servicepresent →<service>-<env>[-<resource-slug>]- Otherwise →
<source_skill-suffix>-<resource-slug>
If the file already exists for the same day, append -HHMM:
<YYYY-MM-DD>-<HHMM>-<scope_slug>.mdExamples:
2026-04-21-data-ai-studio-backend-staging.md2026-04-21-1530-incident-INC-1234-checkout-web.md2026-04-21-logs-data-ai-mcp-timeout-pattern.md
Step 4 — Write the Markdown artefact
Frontmatter contract (no wikilinks inside YAML — keep values as plain strings):
---
title: <short human title — e.g. "data-ai-studio-backend staging latency p95 spike">
created: <YYYY-MM-DD HH:mm>
source: datadog
source_skill: <source_skill>
env: <staging|prod|local|other>
service: <service-or-empty>
project: <inferred-project-or-empty>
resource: <resource-or-empty>
severity: <info|warn|error|critical>
status: open
incident_id: <dd-incident-id-or-empty>
tags: [ops, datadog, triage]
---Frontmatter never contains
notebook_urlornotebook_status— those fields are forbidden by the Hard Constraints.
Body skeleton:
# <Short human title>
**Source skill:** `<source_skill>`
**Scope:** `<service>` / `<env>` / `<resource>`
**Severity:** `<severity>`
**Datadog deep-links (live):**
- [<label>](<url>)
- [<label>](<url>)
- ...
**Created:** <YYYY-MM-DD HH:mm>
---
## Executive Summary
<from findings.Executive Summary>
---
## Evidence
| # | Finding | MCP Tool | Query / Params | Datadog Link |
|---|---------|----------|----------------|--------------|
| 1 | ... | `aggregate_spans` | `service:foo env:staging` | [traces](https://app.datadoghq.com/apm/traces?...) |
| 2 | ... | `analyze_datadog_logs` | `SELECT ... FROM logs WHERE ...` | [logs](https://app.datadoghq.com/logs?...) |
> Every row of the evidence table MUST have a clickable Datadog Link. If the MCP returned data with no natural UI URL (e.g. service-catalog metadata), put `n/a` and explain in the row.
---
## Timeline
<from findings.Timeline, if provided>
---
## Root Cause
<from findings.Root Cause>
---
## Recommendations
### P1 — Immediate
- ...
### P2 — Short-term
- ...
### P3 — Long-term / Follow-up
- ...
---
## Query Audit Trail
| # | Timestamp | MCP Tool | Arguments (summary) | Purpose |
|---|-----------|----------|---------------------|---------|
| 1 | HH:MM:SS | `search_datadog_services` | `team=data-ai-agents` | Locate owned services |
| ... | ... | ... | ... | ... |
**Total queries executed:** X
---
## Next Steps
<from findings.Next Steps — see datadog-triage SKILL for the master-report template; sub-skills typically contribute 2–4 bullets here>
---
## Related
- [[Observability]]
- [[Datadog]]
- [[<project-MOC-if-known>]]
<!-- if related_artefacts was passed, append one bullet per artefact as an unquoted Obsidian wikilink -->Important — wikilinks in frontmatter: never put [[wikilinks]] inside YAML. Keep frontmatter values as plain strings; wikilinks live only in the ## Related body section.
Step 5 — Embed the curated datadog_links
If the caller passed a datadog_links list, render it as the bullet list immediately under the **Severity:** line in the header (template above). This is the user's primary launching pad back into the live Datadog UI from inside Obsidian.
If the caller did not pass datadog_links, mine the Evidence table for the most informative URLs (typically the APM service page, the metric explorer, the logs explorer) and surface them in the header anyway. Aim for 3–6 links — too many becomes noise.
Step 6 — Report to caller
Return to the calling skill:
{
"markdown_path": "<absolute path>"
}And print a brief 2-line human summary:
Published:
MD: <path>Datadog UI URL recipes
Use these patterns when you need to construct deep-links from MCP data (the orchestrator and sub-skills should already have these from their own queries, but include here as a reference):
- APM service:
https://app.datadoghq.com/apm/services/<service>?env=<env> - APM traces (filter):
https://app.datadoghq.com/apm/traces?query=<urlencoded>&historicalData=true - APM single trace:
https://app.datadoghq.com/apm/trace/<trace_id> - Logs explorer:
https://app.datadoghq.com/logs?query=<urlencoded>&from_ts=<ms>&to_ts=<ms> - Logs Flex tier: same with
&storage=flex_tier - Logs patterns view: same with
&viz=pattern&clustering_pattern_field_path=message - Metric explorer:
https://app.datadoghq.com/metric/explorer?query=<urlencoded metric query> - Metric summary:
https://app.datadoghq.com/metric/summary?metric=<metric_name> - Dashboard:
https://app.datadoghq.com/dashboard/<id> - Monitor:
https://app.datadoghq.com/monitors/<id> - Event explorer:
https://app.datadoghq.com/event/explorer?query=<urlencoded> - Incident:
https://app.datadoghq.com/incidents/<id>
URL-encode values that contain :, spaces, { }, etc. (%3A, %20, %7B, %7D).
Promotion Hints
After publishing, check whether the {{DATADOG_TRIAGE_DIR}} folder already contains another triage note with the same service + similar resource/symptom. If yes (≥ 2 occurrences), suggest (do not auto-run) promoting to a runbook using the create-runbook skill. Example hint to the user:
I found a related triage note from
<date>with the same symptom on<service>. Consider promoting both to a runbook with:create a runbook from <path-1>, <path-2>.
If severity is critical and incident_id is set, suggest writing the post-mortem under {{INCIDENTS_DIR}} instead.
Best Practices
- Markdown only — no Datadog notebook is created. Knowledge lives in Obsidian. Re-read the Hard Constraints block at the top of this skill if in doubt.
- Deep links over screenshots — every finding must include a Datadog URL the user can click (traces_explorer_url, trace_deep_link_url, metric explorer, dashboard links, etc.).
- Curated header links — always surface 3–6 live UI URLs in the header so the user has a fast launching pad to the live data.
- Redact secrets — tokens, passwords, customer PII must never make it into the MD.
- Query audit trail is non-negotiable — every query executed during triage must appear in the trail, with arguments summarised (not raw dumps).
- Atomic artefacts — a triage note's evidence and analysis are frozen after creation; the only permitted edits are
status:and a section appended at the bottom (Resolution, Promoted to runbook), per docs/triage-architecture.md. If new evidence arrives, write a new triage note and cross-link. - Never put wikilinks in frontmatter — they break Dataview in this vault. Always body-only.
Edge Cases
- MCP tool fails mid-investigation: write what you have, add a
## Publication Notessection explaining what's missing. - User has no Obsidian configured: the placeholder falls back to
./docs/datadog_analyses/triage/relative to CWD. Warn once. - Caller didn't provide findings in the expected shape: do NOT fabricate content. Ask the caller to re-emit findings in the contract shape.
- Scope lacks service: use
source_skillas fallback in the scope_slug (e.g.2026-04-21-logs-<resource>.md). - Caller passes
notebook_cells(or any Datadog write directive): silently drop the field, log in## Publication Notes, and emit only the MD. Do NOT call any Datadog write tool.
Dependencies
Required:
- Obsidian vault configured via
install.sh --reconfigure(optional; otherwise falls back to./docs/...).
Optional:
create-runbookskill (for promotion suggestions after ≥ 2 related triages).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

