justinmclean opened a new pull request, #253:
URL: https://github.com/apache/airflow-steward/pull/253
> **Generated by the spec-driven build loop.** This page was produced by an
> autonomous run of `tools/spec-loop` (`./loop.sh` — one work item, one
branch,
> one PR). Authored by Claude (see the `Generated-by` commit trailer) and
> reviewed + tested by a human before submission.
## What
Adds `docs/mode-economics.md` — an indicative per-skill / per-mode /
per-model
token-cost page, so a maintainer evaluating adoption can see the cost shape
up
front rather than discovering it after the fact. Covers Triage, Mentoring,
Drafting, Pairing and Auto-merge; model-class trade-offs; local/self-hosted
inference; cost-reduction levers; and the long-term ASF inference-endpoint
roadmap item. Adds a reciprocal cross-reference from `docs/modes.md`.
## Why
`MISSION.md § Affordability` commits to documenting mode economics honestly.
This page is that documentation, and its data is the quantitative input for
future ASF inference-endpoint capacity planning.
## Changes
- `docs/mode-economics.md` — new page.
- `docs/modes.md` — cross-reference back to it.
## Testing — and a limitation it shipped with
The doc gates runnable locally pass: the doctoc TOC matches all 15 headings,
every
internal link/anchor resolves (including the MISSION § Affordability
deep-link and
the reciprocal `modes.md` cross-ref), file-hygiene is clean, and 12 of 13
referenced
skills exist on `main` (the 13th, `pairing-self-review`, is a sibling
unmerged
branch — a name in a table, not a link). `markdownlint`/`typos`/`lychee`
should be
confirmed by CI.
**Limitation found in testing — the loop did not detect it.** The page's
headline
anchor claimed "one skill file (SKILL.md) ≈ 3,000–6,000 tokens" and that this
overhead "applies" generally. Tokenizing the real catalogue with
`cl100k_base`
showed that holds only for the *median* skill: of 26 skills, **9** fall in
3k–6k,
**6** are below, and **11** are above — the large multi-step security skills
run far
higher (`security-issue-import` ~22,000, `security-issue-sync` ~36,000, i.e.
4–6×
the stated ceiling). The anchor is corrected here to measured, class-specific
ranges (small ~1k–3k, typical ~3.5k–9k, large security ~11k–36k). The build
loop
shipped a plausible but unverified figure; the test-before-PR step is what
caught it.
## Notes
- Indicative cost guidance; the counts in the "what tokens means here"
anchor are
now measured (`cl100k_base`), not estimated.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
