[Bug]: Template files are inconsistently formatted

templates/ 内の 5 つのファイルを、create-new-feature.sh、create_new_feature.py、PowerShell 版、および templates/commands/specify.md と併せて比較します。まず、どの不一致が意図的なものかを判断し、標準となるフィールド名、レイアウト、front matter、置換について合意します。合意した規約が一貫して適用され、生成されたテンプレートの内容に意図しないプレースホルダーが残らなくなれば完了です。

Bug Description

The speckit template files are inconsistently formatted.

Steps to Reproduce

1. Clone the latest release and enter it:

   git clone --depth 1 --branch v1.0.1 https://github.com/github/spec-kit.git
   cd spec-kit
   

2. Read the five templates in templates/:

   for f in spec plan tasks checklist constitution; do
     echo "===== $f-template.md ====="; cat "templates/$f-template.md"
   done
   

Version checked: v1.0.1, tagged 2026-08-21. templates/ tree hash 8730ddb.

Expected Behavior

Consistent templates.

Actual Behavior

Inconsistent templates.

Specify CLI Version

1.0.0

AI Agent

Claude Code

Operating System

Not relevant

Python Version

Not relevant

Error Logs

Additional Context

I've been working with the stock templates and noticed a handful of small inconsistencies between them. Individually none of these matter much, but I think this is a place where consistency pays off more than usual: these templates are the standard the model follows when it fills them in. If the standard is ambiguous, the output inherits the ambiguity. So I'd rather raise them than not.

I looked at spec-template.md, plan-template.md, tasks-template.md, checklist-template.md, and constitution-template.md.

Issues

The pointer to the upstream document has three names and three formats. Three templates each point at the document they derive from, and no two agree on what to call it:

Template	Field	Value
plan-template.md:5	Input	Feature specification from /specs/[###-feature-name]/spec.md
tasks-template.md:8	Input	Design documents from /specs/[###-feature-name]/
checklist-template.md:5	Feature	[Link to spec.md or relevant documentation]

plan-template.md names it twice, under two names, in two formats: line 3 carries **Spec**: [link] (an unfilled placeholder) and line 5 carries the Input field above (a backticked path to that same document).

The values also differ in mechanism, not just wording: two are literal paths containing a [###-feature-name] placeholder, one is a bare [Link to ...] placeholder. Nothing states which is expected.

spec-template.md uses Input for something else. Line 9 is **Input**: User description: "$ARGUMENTS" — the raw user request, substituted by the command, not a pointer to an upstream document. That may be a reasonable use of the word, but it means Input denotes two unrelated things across the template set: provenance in plan/tasks, captured arguments in spec.

The date fields could be more precise. spec-template.md and checklist-template.md use Created; plan-template.md uses Date, which doesn't say whether it means created or last-edited. Date Created and Date Last Edited would be unambiguous, and would let a template carry one, the other, or both as appropriate.

Some of these fields could be filled mechanically, and currently are not. $ARGUMENTS shows the templates already support machine substitution, but the values that are mechanically derivable are left to the model:

• [DATE] is never substituted. create-new-feature.sh:362 copies the spec template through verbatim (printf '%s' "$SPEC_TEMPLATE_CONTENT" > "$SPEC_FILE") with no rewriting of the body. The only date logic in the scripts (create-new-feature.sh:263, create_new_feature.py:276, and the PowerShell equivalent) generates a timestamp for the feature number in a fallback path — it never touches [DATE]. So the creation date reaches the model as a literal placeholder for it to fill in. templates/commands/specify.md:152 emits the same **Created**: [DATE] into a generated checklist, reproducing the gap downstream.
• The upstream pointer is always derivable. The Input of plan.md is always the spec.md of the same feature.
• Could [FEATURE NAME] be substituted too? create-new-feature.sh already computes BRANCH_NAME and FEATURE_NUM and returns them on stdout, so a feature identifier exists at generation time. I have not looked into whether the name itself is available at that point, or whether leaving it to the model is deliberate.

Status exists in one template only, with undefined values. spec-template.md ships Status: Draft. Nothing I could find defines the other legal values, and no other template carries the field. I'm unsure whether the intent is that only specs are status-tracked, or whether it's meant to be broader.

Header layout varies, including within one file. Most templates put one field per line. plan-template.md pipe-joins three fields on its first metadata line and then switches to one-per-line.

tasks-template.md is the only template with YAML front matter, and there's a blank line between the opening --- and the description: key, which I believe stops most parsers from reading it as front matter at all. Since no other template has front matter, I wasn't sure whether it's needed here. Either way I think it might not be a bad idea for all front matter to be YAML, but I assume you have already considered this.

Less important

Instructional comments use a few different conventions. Some blocks are marked ACTION REQUIRED, others are boxed IMPORTANT … SAMPLE blocks, and constitution-template.md uses inline <!-- Example: -->. The Example: usage reads as genuinely different to me — it labels illustrative values rather than flagging content to delete — but the first two seem to overlap.

Sample content isn't flagged uniformly. tasks-template.md and checklist-template.md both say explicitly not to keep the sample items. spec-template.md ships three sample user stories without an equivalent instruction, though they look like the same kind of placeholder.

Probably deliberate

A couple of differences look deliberate rather than accidental, and I'd leave them alone: constitution-template.md reasonably has no branch, input, or status fields since it isn't per-feature, and its version/ratified footer is a document lifecycle rather than header metadata.

Caveats

Happy to be told some of these are intentional. I may be missing context on how the commands consume these fields.

The field names, values, and line numbers above are quoted from the v1.0.1 tag, and the claim that [DATE] is never substituted was checked against the scripts. The reasoning about intent is mine and may be missing context. I don't have time to spend more time on this without knowing that the issue is something you think is important enough to dive into. If you agree that this is a real issue I could sit down and do the actual work after we decide what the appropriate solution should be.

Possible long-term solution

Perhaps the long term solution should even be to create some kind of template generator. This would allow mechanically verifiable consistency and the possibility for users to change the formatting of templates if they want to, or create new ones which might be needed for extensions. For example, I got started with this review because I prefer links in my documents to be proper markdown links so they can be followed by humans.