Skip to content

dimagi-internal/ace / skills/sweep-ace-web

sweep-ace-web

Verified

Bulk-delete uploaded chat sessions from a deployed ace-web, after a report and your approval.

sweep-ace-web

Bulk wipe, not orphan sweep. Lists every Session in workspaces where the caller is Owner or Editor on the deployed ace-web, prints the plan grouped by workspace, prompts the human, then deletes all approved ids in one call. CASCADE on the ace-web side handles IngestUpload, Message, SessionParticipant, ShareToken, and Draft rows.

This skill does not consult the live-set. The user has explicitly opted into wiping all visible sessions; opp-orphan filtering is not its job.

Inputs

  • ACE_WEB_BASE_URL — env var; deployed ace-web base URL (no trailing slash). Defaults to https://labs.connect.dimagi.com/ace to match --ace-web-url elsewhere in the plugin.
  • ACE_WEB_PAT_TOKEN — env var; ACE's own ace-web token by default (written by /ace:setup), or yours if you minted one with /ace:ace-web-pat-mint. The sweep is scoped to whichever workspaces that identity belongs to on the ace-web side.
  • sweepFolder — timestamped sweep folder created by the orchestrator (e.g. ACE/_sweep/<timestamp>/). Products land here.
  • mode — recommend (default) | execute. In recommend the skill is report-only: list + render the plan + return it; no delete call. In execute it deletes only approvedIds. The human-confirmation gate between the two is the orchestrator's job — see agents/sweep.md § Human-confirmation gate.
  • approvedIds (execute mode only) — the exact session ids the human approved in chat. Delete nothing outside this set.

No liveSetPath input. If the orchestrator built one for a sibling sweep, ignore it.

Products

  • <sweepFolder>/ace-web-sessions.md — human-readable plan grouped by workspace, with totals at the top.
  • <sweepFolder>/ace-web-sessions.yaml — machine-readable session list (for replay).
  • <sweepFolder>/ace-web-sessions-result.yaml — written post-execution with deleted count and failed[] rows (if any).

Process

  1. Verify preconditions.

    • ACE_WEB_PAT_TOKEN set; if missing, stop and instruct the operator to run /ace:ace-web-pat-mint.
    • ACE_WEB_BASE_URL resolved (env var or default). Strip any trailing slash.
    • sweepFolder provided by the orchestrator. If absent, abort — the skill must not invent its own.
  2. GET <ACE_WEB_BASE_URL>/api/sessions/sweep with -H "Authorization: Bearer $ACE_WEB_PAT_TOKEN". Expect 200. Response shape:

    sessions:
      - id: int
        slug: str
        title: str
        source: "web" | "upload"
        status: "active" | "archived" | "imported"
        opp_slug: str            # may be ""
        opp_run_id: str          # may be ""
        workspace_slug: str
        message_count: int
        upload_count: int
        created_at: ISO-8601
        updated_at: ISO-8601
    total_raw_bytes: int
    

    On 401, instruct the operator to re-mint the PAT. On any other non-200, print the body and abort.

  3. Render the report. Group rows by workspace_slug. Per-workspace section shows: row count, source breakdown (upload vs web count), opp linkage breakdown (with-opp vs unlinked count), and a short table of the first 30 rows for that workspace (slug, source, opp_slug, title, updated_at). Headline: "ace-web sweep plan — N sessions across M workspaces; T MB of raw transcripts." drive_create_file to <sweepFolder>/ace-web-sessions.md and the YAML peer ace-web-sessions.yaml. The YAML must contain every row in full (not truncated) so a replay knows exactly which ids to send.

  4. Recommend — stop here in mode: recommend. Return the plan to the orchestrator: the full session-id list (this is a bulk wipe — every session the PAT can write to, not an orphan-filtered subset; say so plainly so the human understands the scope) + per-workspace counts + the report Drive link + the reversibility note (delete CASCADEs through IngestUpload, Message, SessionParticipant, ShareToken, Draft — not reversible). Issue no delete call. Do not try to prompt the human from this skill — a dispatched subagent can't reach them; the orchestrator runs the confirmation gate (see agents/sweep.md § Human-confirmation gate). If the human declines at that gate, the orchestrator never dispatches the execute pass.

Execute phase (mode: execute only)

Runs only when the orchestrator re-dispatches with mode: execute + approvedIds (the session ids the human approved in chat — typically all, but honor a subset if that's what was approved).

  1. POST <ACE_WEB_BASE_URL>/api/sessions/sweep/delete with -H "Content-Type: application/json", -H "Authorization: Bearer $ACE_WEB_PAT_TOKEN", and {"session_ids": [<approvedIds>...]} as the body. Expect 200. Response shape:

    deleted: int
    failed:
      - session_id: int
        reason: "not_found" | "forbidden" | "db_error: <msg>"
    

    Write the response to <sweepFolder>/ace-web-sessions-result.yaml with decision: "approved" for downstream summary.

  2. Report. Return a short summary block:

    ace-web sweep — N approved, K deleted, F failed
      deleted: K sessions across <M> workspaces
      failed:  F items (see ace-web-sessions-result.yaml)
    

Shell reference

[ -n "$ACE_WEB_PAT_TOKEN" ] || { echo "ACE_WEB_PAT_TOKEN not set; run /ace:ace-web-pat-mint"; exit 2; }
BASE_URL="${ACE_WEB_BASE_URL:-https://labs.connect.dimagi.com/ace}"
BASE_URL="${BASE_URL%/}"

# Scratch response files: `mktemp`, never a fixed /tmp literal (ace#1046) —
# /tmp is shared across macOS users, so `-o` can fail EACCES while the
# follow-up `cat` silently returns another session's response body.
LIST_JSON="$(mktemp "${TMPDIR:-/tmp}/ace-sweep-list-XXXXXX.json")"
DEL_JSON="$(mktemp "${TMPDIR:-/tmp}/ace-sweep-del-XXXXXX.json")"

# 1. List
HTTP=$(curl -sS -o "$LIST_JSON" -w '%{http_code}' \
  -X GET "$BASE_URL/api/sessions/sweep" \
  -H "Authorization: Bearer $ACE_WEB_PAT_TOKEN")
[ "$HTTP" = "200" ] || { echo "list $HTTP"; cat "$LIST_JSON"; exit 3; }

# 2. (Render + approval handled by the agent calling drive_create_file)

# 3. Delete — agent constructs IDS from approved list
HTTP=$(curl -sS -o "$DEL_JSON" -w '%{http_code}' \
  -X POST "$BASE_URL/api/sessions/sweep/delete" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $ACE_WEB_PAT_TOKEN" \
  -d "{\"session_ids\": [$IDS]}")
[ "$HTTP" = "200" ] || { echo "delete $HTTP"; cat "$DEL_JSON"; exit 4; }

Failure modes

HTTPCauseRemedy
401ACE_WEB_PAT_TOKEN missing/revokedRe-run /ace:ace-web-pat-mint.
422Body shape invalid (non-int ids)Verify the YAML round-trip; ids must be JSON integers.
200 with failed[] populatedPer-row failures (e.g. session removed between list and delete, or workspace role changed mid-sweep)Treat not_found as success-equivalent for idempotency. Re-run on db_error. forbidden means the user's role changed during the sweep.

Implementation notes for agents

  • Session is the only ORM aggregate root. The ace-web endpoint relies on Django's on_delete=CASCADE to drop IngestUpload, Message, Draft, SessionParticipant, and ShareToken rows. Do not attempt to delete those individually.
  • Workspace scoping happens server-side via the PAT's user's WorkspaceMembership rows. The skill never sends a workspace filter — the server already knows what the caller can write to.
  • The list endpoint is unpaginated by design. If a deployment ever accumulates enough sessions to make a single 200 unwieldy, surface that back to the ace-web team rather than papering over it client-side.
  • total_raw_bytes is summed from IngestUpload.raw_bytes server-side. Display as MB in the report headline (bytes / 1048576).

Related skills

  • sweep-opp-runs is the other sweep that ignores the live-set (retention, not orphan). Same shape: list → render → approve → execute.
  • upload-transcript is the inverse atom — it's how the sessions wiped by this skill got there in the first place.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 319kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 118kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall