Role
You record what was implemented for a task as a structured comment on the Azure DevOps work item. The log is searchable via ADO work-item search / WIQL and serves as the canonical knowledge base for "what exists already" — future engineers and AI agents should search the board before creating new code.
You are the Azure DevOps parallel of jira-impl-logger / linear-impl-logger. The
workflow (and task-executor / sprint-executor via the adapter's implLogger)
routes here when tracker: ado. You are invoked once per completed task, or directly
to backfill a log.
Why a work-item comment and not repo files: repo implementation logs decay or duplicate the git log. ADO work-item comments are already the system of record for work done, are searchable, and link naturally to the work item's full context (description, links, state, iteration).
Inputs
- Work item id (e.g.,
1234) - Summary — one-line description of what was implemented
- Files created — list of paths
- Files modified — list of paths
- Statistics — linesAdded, linesRemoved (from
git diff --shortstat) - Artifacts — structured data about what was built (see schema below)
- Commit hash(es) — for back-reference
- Optional: docs backend —
docs: confluence | notion | ado-wiki | local(from workflow-config) enables the docs mirror, plus the Impl Logs dataset/page id
Artifact Schema
Structured data, adapted from spec-workflow-mcp. Each artifact type (apiEndpoints, components, functions, classes, integrations, dataModels, pipelineStages) is optional — include the ones relevant to this task.
The schema is defined once, shared verbatim across all four impl-loggers — see prompts/coding/templates/impl-log-artifact-schema.md.
Workflow
Step 1 — Validate Inputs
- Confirm the work item exists (Azure DevOps MCP
wit_get_work_item) - Confirm at least one artifact category is provided — empty artifacts is a failure mode
- Confirm files lists are non-empty (a task that created/modified no files is suspicious)
If inputs are thin, ask the caller for more detail before posting. A sparse log is worse than no log — it pollutes search results.
Step 2 — Render the Comment
Use the same structure as prompts/coding/templates/jira-impl-log-template.md:
## Implementation Log — AB#{id}
**Summary:** {one-line summary}
**Commit:** {hash} — {commit message first line}
**Date:** {ISO date}
### Files
- Created: {file1}, {file2}
- Modified: {file3}, {file4}
- Stats: +{linesAdded} / -{linesRemoved} across {N} files
### Artifacts
#### API Endpoints
| Method | Path | Purpose | Location |
| ... | ... | ... | ... |
#### Components
| Name | Type | Purpose | Location | Props |
#### Functions
| Name | Purpose | Location | Signature | Exported |
#### Classes
| Name | Purpose | Location | Methods | Exported |
#### Integrations
| Description | Frontend | Backend | Data flow |
#### Data Models
...
#### Pipeline Stages
...
### Searchable Keywords
{comma-separated list of class/function/endpoint names for ADO search}
Step 3 — Post Comment
Use wit_add_work_item_comment(project, workItemId:id, comment:rendered, format:'markdown').
Step 4 — Tag the Work Item
Add tag impl-logged via wit_update_work_item(id, {"System.Tags": "{existing}; impl-logged"}). This lets queries/dashboards filter work items that have logs vs. those still missing them.
Step 5 — Mirror to the docs surface (when a docs backend is configured)
If docs is configured, also publish the log to the docs surface via the docs adapter's publishRecord (skills/feature-spec-author/references/docs-adapter.md):
- docs=ado-wiki → append the rendered log to the Impl Logs wiki page (a row in its table) under
{root}/ImplLogs. - docs=notion → a row in the Impl Logs database; body = the rendered markdown.
- docs=confluence → append the rendered log to the Impl Logs page (or a row in its table macro).
- docs=local → append to
documentation/impl-logs/{id}.md. - If the Impl Logs surface doesn't exist yet, do not silently skip — tell the caller to provision it. The work-item comment (Step 3) is the system of record regardless.
Step 6 — Return
Return to caller:
- ADO work-item comment URL
- Confirmation that the work item is tagged
- Docs mirror URL (if mirrored)
Why Structured Artifacts Matter
Future AI agents will query the ADO board (work-item search / WIQL) before writing new code. Incomplete logs produce duplicate endpoints, duplicate components, duplicate helpers.
Minimum quality bar:
- If the task created an API endpoint,
apiEndpointsMUST be populated - If the task created a UI component,
componentsMUST be populated - If the task created a class,
classesMUST be populated - A pure refactor (no new public surface) may have empty artifacts but MUST explain in summary: "Refactor only — no new public surface"
Backfilling Logs
If invoked on a closed work item that never got a log:
- Read the work item description and commits referenced
git log --followandgit showto identify files and stats- Infer artifacts from the diff (grep for
def,class,@app.route, React component exports) - Offer the draft log to the user for confirmation before posting
What This Skill Does NOT Do
- Does not evaluate code quality (that's
clean-code-reviewer) - Does not close the work item (that's
sprint-executor) - Does not write logs to the repo (deliberately — the work item is the log surface)
- Does not add commits or run tests
References
skills/task-executor/references/tracker-ado.md— the ADO tracker adapter (read this first for tool names).skills/jira-impl-logger/SKILL.md·skills/linear-impl-logger/SKILL.md— the JIRA / Linear parallels (same artifact schema).skills/feature-spec-author/references/docs-adapter.md— the docs adapter (Step 5, docs mirror).prompts/coding/templates/jira-impl-log-template.md— the log content template.- Azure DevOps MCP:
wit_add_work_item_comment,wit_update_work_item,wit_get_work_item. - Origin schema: https://github.com/Pimzino/spec-workflow-mcp
log-implementationtool
Feedback
If the user corrects this skill's output due to a misinterpretation or missing rule in the skill itself (not a one-off preference), invoke skill-feedback to capture structured feedback and optionally post a GitHub issue.
If skill-feedback is not installed, ask the user: "This looks like a skill defect. Would you like to install the skill-feedback skill to report it?" If the user declines, continue without feedback capture.