Multica Goal Tracker

Multica Goal Tracker

Use this skill to keep Multica issues consistent when the human workflow is:

1. Create or reuse a Multica issue.
2. Paste the /goal command that will drive a long $intuitive-flow or

related skill run.

1. After the run completes, attach concise completion evidence.
2. If the result is incomplete, keep the issue open and append the next goal as

another attempt on the same issue.

1. Repeat until the issue-level outcome is complete. Finish comments show the

per-goal attempt plus cumulative issue time.

The skill can also create the Multica issue from an approved $intuitive-preflight contract. That path records planning provenance and the exact executable /goal, then immediately appends the normalized tracker start comment. It does not execute the goal and does not create finish evidence.

Prefer appending comments over rewriting issue descriptions. Only update the description when the user explicitly wants the issue top-level description cleaned up.

All comments created by this skill use the current Multica user token, but they must begin with the visible marker:

> Agent 提交：以下内容由 Agent 帮忙整理并提交，用于和人工手写评论区分。

Keep this marker as the first visible line on start, finish, and final-review comments.

Create From Preflight

Use this after an $intuitive-preflight contract is approved and before the goal is executed. The preflight must be DRAFT/approved, not BLOCKED_NEEDS_DECISION, and it must contain either Main-session /goal prompt or To execute with the executable /goal.

bun skills/multica-goal-tracker/scripts/track_goal.ts \
  create-from-preflight \
  --preflight-file /tmp/preflight.md \
  --workspace-id roboclaws

Never rely on the Multica CLI default workspace when creating an issue from a preflight. The default workspace may point at another product. If the user gives a URL such as https://multica.evad.mioffice.cn/roboclaws/, pass --workspace-id roboclaws or the full URL so the script can resolve and verify the real workspace id before writing.

Keep create-from-preflight issue descriptions intentionally short. Do not paste the full ## Preflight Contract into the issue body; the plan file is the source of truth for detailed scope, non-goals, context, and verification. Humans should be able to scan the issue quickly, then open the plan only when they need the full contract.

Use --dry-run first when validating a new preflight shape. Run bun skills/multica-goal-tracker/scripts/track_goal.ts --help for the current option surface.

Do not use create-from-preflight for conversation-only work unless the preflight body contains the full approved contract; otherwise context compression can erase the issue's source of truth. Do not treat the created issue or start comment as completion evidence. Finish evidence still must come from a real run/session as described below.

Start Tracking

Use this after the issue exists and contains the goal, or when the user gives you the goal text directly. Run the command from the Intuitive Flow checkout root so the script path stays portable across clone locations.

bun skills/multica-goal-tracker/scripts/track_goal.ts \
  start \
  --issue MIA-40

If the goal is not already in the issue description, pass --goal-file.

bun skills/multica-goal-tracker/scripts/track_goal.ts \
  start \
  --issue MIA-40 \
  --goal-file /tmp/goal.txt

Pass --update-description only when the user asks to normalize the issue description itself. The script inserts or replaces a marked summary block and preserves the existing description below it.

Use --dry-run before writing to a real issue when testing a prompt or script change.

Finish Tracking

Use this after the goal run is complete or after the user asks to preserve the state of a goal attempt on an existing issue. Do not create a new issue for this flow unless the user explicitly asks. Finish evidence must come from real session output: Multica execution run messages by default, a Codex JSONL session, a skill-runner run directory, or an explicit transcript via --session-file.

bun skills/multica-goal-tracker/scripts/track_goal.ts \
  finish \
  --issue MIA-40

The script picks the active goal, extracts real run/session evidence, records attempt metadata, and posts one scannable comment with overview, timeline, details, and raw selected output. complete attempts are completion records; partial, blocked, and failed attempts are execution records.

If the issue has no Multica run history, finish fails fast instead of creating a fake proof card. In that case pass a real Codex session JSONL for the finished attempt:

bun skills/multica-goal-tracker/scripts/track_goal.ts \
  finish \
  --issue MIA-40 \
  --session-file ~/.codex/sessions/2026/06/04/rollout-....jsonl

Or pass a real skill-runner run directory. It uses compact artifacts and avoids noisy terminal.log output.

bun skills/multica-goal-tracker/scripts/track_goal.ts \
  finish \
  --issue MIA-40 \
  --session-dir ~/.cache/skill-runner/runs/

Use --allow-manual-summary --summary "..." only when the user explicitly accepts a manual fallback. Manual fallback is not real session proof.

If an attempt is incomplete, finish it with --attempt-status partial|blocked|failed, then run start again for the follow-up goal. The next finish becomes the next attempt and the card shows cumulative issue time.

Final Review

Use this when the human wants one final review thread for an issue that already has multiple goal attempts, especially when an earlier attempt was partial and a follow-up goal completed the issue. Do not hand-compose the comment. Put the attempt data in JSON and let the script own the format.

bun skills/multica-goal-tracker/scripts/track_goal.ts \
  final-review \
  --issue MIA-40 \
  --attempts-file /tmp/multica-goal-attempts.json

final-review posts one text comment with a short issue-level summary, compact attempt list, overview, timeline, details, and complete raw outputs for each attempt. The comment stores metadata for every attempt, so later tracker runs can recover cumulative duration even if older Agent comments are cleaned up.

Useful Options

Use --help for full syntax. The high-value knobs are --workspace-id, --goal-file, --session-file, --session-dir, --run-id, --attempt-status, --proof, --dry-run, and --allow-manual-summary.

Style

Default generated issue text to Chinese. Keep summaries short. The issue should answer "what is this goal trying to accomplish?" without becoming a plan document. Preserve the raw /goal block because it is the execution provenance.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: MiaoDX
• Source: MiaoDX/intuitive-flow
• License: MIT

Install and usage instructions live in the source repository linked above.