agentleFS
seankoji-comseankoji-com/.agents/skills/plan-report/SKILL.md

plan-report

Skill2 starsChanged 3 days agoMITView on GitHub

Propose a project plan or report progress and final calibration as a Gantt-style HTML report from a private JSON ledger. Use for coordinated runs, explicit timeline requests, or when ship, backlog-burn or implement calls for plan reporting.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 4 are distinctive, and 2 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.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the ledger contract before creating or changing a ledger.

    line 29
  2. Distinctive:

    Before asking a material owner question, check the latest decision record and settled answers.

    line 49
  3. Distinctive:

    Read the real clock at render time and set as-of to that observation in Australia/Sydney with its correct offset.

    line 61
  4. Very specific:

    Update statuses, forecasts, observed segments, review rounds/results, decisions and events.

    line 64
  5. Distinctive:

    Run the renderer using absolute input/output paths, from any working directory:

    line 77
  6. Very specific:

    Use --mode propose or --mode final for those boundaries.

    line 84

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Show the plan and its progress

The coordinator keeps one JSON ledger per project and renders it with the bundled Python 3.10+ standard-library helper. The ledger records estimates, observations and decisions; it does not dispatch work or replace the host's task panel.

Cadence and ownership

  1. Render at proposal, task completion, final handback and on request, including solo tasks. Task completion means the agreed endpoint, such as merge or close, not each review or fix round; those get short chat updates. Apply an adopted project's report_cadence override when present. Whole-run timelines appear only at proposal and final handback; progress reports show the recent timeline.
  2. The main-thread coordinator is the sole ledger writer and presenter. Delegates return segment start/completion timestamps, actual model/provider/effort, review round/result and exposed usage with sources. They never present the report. Use host start/completion notifications when available; label uncertain timing rather than inventing precision.
  3. Reuse the caller's authority and settled decisions. Plan approval freezes estimates; it adds no authority to publish, dispatch, spend or change scope.

Propose

  1. Read the ledger contract before creating or changing a ledger. Read host cost evidence when collecting usage or choosing its scope; refresh capability claims against the current host.
  2. Choose a repository-ignored work directory inside the session's recorded worktree and verify the ledger and output paths with git check-ignore before writing. Keep ledgers and reports untracked, private and out of GitHub. Removing the worktree loses this history; use another destination only with authority. Checked-in fixtures must be synthetic.
  3. Record project name/status, Australia/Sydney timezone, run start and as-of; tasks, streams, dependencies, owner identities, forecasts, build/review allocations and planned review rounds; milestones/gates with required reviewers; decision records and relevant external events. Estimates are wall-clock minutes, separate from tokens, turns and USD.
  4. Render proposal mode and present it. Record actual owner approval, its timestamp and approver. Freeze each task's original baseline at that point. A proposal or silence is not approval. Continue only work already authorised while a decision is pending.

Report

Before asking a material owner question, check the latest decision record and settled answers. Batch foreseeable questions through the host-native question or input tool, with the recommendation first. Continue independent authorised work while an answer is pending. Rendering a report is not asking the question.

Record the actual prompt, owner, observed asked time and affected tasks as pending. On a real reply, append a decided snapshot with the actual answer, observed decision time and optional rationale/source, keeping earlier records. Unknown times stay unknown. A recommendation or silence is not a decision. Do not re-prompt every progress turn or ask again after an answer is recorded. The coordinator owns asking and recording replies; the renderer only displays the ledger.

  1. Read the real clock at render time and set as-of to that observation in Australia/Sydney with its correct offset. Never advance it to a forecast or imagined task completion. Historical observation boundaries remain valid. Update statuses, forecasts, observed segments, review rounds/results, decisions and events. Record each change of model or effort as a new segment. Refresh forecast.end for every running/review task on every coordinator update, even when no report is due; never invent an ETA to hide a stale forecast. Open segments end at as-of for the current view. Mark waiting periods as scoped events; do not silently charge them as build/review time.
  2. Copy cost only from exposed host observations. Preserve the source and whether USD is a host estimate. Missing data is unknown, including unavailable USD on subscriptions. Keep each metric at either task or segment scope to avoid double-counting. A task total spanning models remains unattributed.
  3. Preserve original baselines. A replan appends a timestamped rebaseline with reason, approver and complete changed-task allocations. Update forecasts separately. Earlier estimates and overruns remain visible.
  4. Run the renderer using absolute input/output paths, from any working directory:

    python3 /installed/plan-report/scripts/render.py /worktree/ignored/ledger.json \
      --mode report --output /worktree/ignored/report.html

    Use --mode propose or --mode final for those boundaries. --theme light, dark or auto controls appearance; --recent-hours 4 shows four hours before and after as-of. Recent comes first and stays expanded; each later section has its own closed caret when it contains data. Empty sections are omitted entirely. Whole-run appears only in propose and final, never report, including on-request progress updates. Dates show absolute Sydney local clocks without repeated timezone labels; source offsets and DST arithmetic stay intact. The renderer is a single relocatable file. Invalid ledgers exit nonzero with a useful message; correct the ledger before presenting. Future as-of is rejected against the renderer's clock. Only synthetic replay fixtures may use --allow-future-fixture; their output is visibly marked and cannot serve as live evidence. Past fixtures need no exception.

  5. In T3 the main thread uses html_preview then html_render on the generated file. Outside T3 use the host's equivalent artifact or an accessible file link. Reports are self-contained and make no network requests. Inspect the timelines included in that mode, scoped event rows, dependencies and populated tables before claiming a report is readable. Explain any missing host presentation capability.

Final calibration

  1. Close observed segments and reconcile completed, cancelled and unfinished work. Render final mode even when the endpoint is incomplete; identify remaining gates.
  2. Compare original approved build/review allocations with completed segments by provider, model, effort and kind. Compare planned and distinct actual review rounds per task. A changed identity makes that task/kind comparison not comparable; keep its original estimate rather than transferring it to the replacement model. Ratios use observed segment minutes, not task span or summed parallel project elapsed time.
  3. Report sourced known cost subtotals and missing coverage separately. Unapproved proposals and unfinished tasks cannot establish final estimate accuracy. Feed this calibration and its limitations into ship's retrospective when ship is the caller. Keep local checks, independent review, hosted CI, publication, deployment and live acceptance separate.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.