Skip to content
Agentmart
Submit a tool

maintain-project-context

Clean, compact, and reconcile durable project context and any configured issue-tracker mirror without changing established meaning.

Skill by toy-crane

IndexedMIT★ 3
Write a review

Chalk Labs

Need help deploying maintain-project-context?

Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.

Book a call

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/toy-crane-maintain-project-context

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Clean, compact, and reconcile durable project context and any configured issue-tracker mirror without changing established meaning. Always use this skill whenever a request cleans, audits, or checks consistency across two or more durable context sources, even when they are called product docs, glossary, decisions, specs, or agent instructions instead of PRODUCT.md, GLOSSARY.md, docs/decisions, docs/specs, AGENTS.md, or CLAUDE.md. Also use for confirmed shipped-spec retirement or a full consistency sweep between default-branch spec folders and their tracker issues. Do not use to define new product intent, record one newly settled term or decision, look up context, shape a feature, implement work, or revise an active spec's behavior against work that has shipped, which belongs to `babysit-specs`.

Use cases: productivity, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md and code. Checked Oct 26, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/maintain-project-context

What's inside

Skill name
maintain-project-context
Description
Clean, compact, and reconcile durable project context and any configured issue-tracker mirror without changing established meaning. Always use this skill whenever a request cleans, audits, or checks consistency across two or more durable context sources, even when they are called product docs, glossary, decisions, specs, or agent instructions instead of PRODUCT.md, GLOSSARY.md, docs/decisions, docs/specs, AGENTS.md, or CLAUDE.md. Also use for confirmed shipped-spec retirement or a full consistency sweep between default-branch spec folders and their tracker issues. Do not use to define new product intent, record one newly settled term or decision, look up context, shape a feature, implement work, or revise an active spec's behavior against work that has shipped, which belongs to `babysit-specs`.
License
MIT
Instructions
186 lines in SKILL.md · view on GitHub

Files (9)

  • SKILL.md
  • openai.yaml · agents/
  • evals.json · evals/
  • PRODUCT.md · evals/fixtures/conflicting-code/
  • team-invitations.md · evals/fixtures/conflicting-code/src/
  • PRODUCT.md · evals/fixtures/settled-audience/
  • README.md · evals/fixtures/settled-audience/docs/decisions/
  • product-audience.md · evals/fixtures/settled-audience/docs/decisions/
  • trigger-evals.json · evals/

SKILL.md (first lines)

# Maintain Project Context

This skill requires `define-product`. Before doing any work, confirm it is
installed. When it is missing, stop and report it with its install command,
`npx skills@latest add toy-crane/skills --skill define-product`, using the
project's package manager runner where it pins one.

Run a deliberate hygiene pass across the durable context that future work will
reuse. Make the current meaning easier to retrieve without becoming the
decision-maker for any artifact.

## Preserve authority

- Apply a meaning only when the user confirmed it, explicitly delegated that
  class of decision, or an authoritative project source already states it.
- Use shipped code to spot stale documentation, not to prove intent. Do not
  change product code during this pass.
- Leave sources unchanged when their intended meaning conflicts or shipment is
  unclear. Ask the exact question required to continue; never choose by file
  date, implementation prevalence, labels such as `accepted`, or silence.
- Preserve a current fact or decision only in the artifact that owns it. Do not
  build a parallel summary or copy the same authority across files.

## Inspect the durable surfaces

Inspect only the surfaces that exist or are required by the cleanup:

- root `PRODUCT.md` for the app-level product premise;
- root `GLOSSARY.md` for current project terms;
- `docs/decisions/README.md` and the relevant subject contracts;
- `docs/specs/<slug>/` folders whose work may have shipped;
- always-loaded repository guidance such as `AGENTS.md` and `CLAUDE.md`.

Do nothing merely because an optional file is absent. This is a periodic
cross-artifact pass, not the incremental capture of a single newly settled term,
decision, or follow-up.

## Maintain product context

When `PRODUCT.md` exists, compare its claims with explicit current product
decisions and other authoritative context. Remove duplicated or obsolete
wording and apply already-settled product meaning while preserving still-current
content.

Keep `PRODUCT.md` product-only: app definition, users and situations, problem
and alternatives, promised change, core loop, app-wide capabilities and
boundaries, experience principles, success signals, assumptions, and unknowns.
Do not move stack, architecture, repository mechanics, screen requirements,
work-unit scope, or acceptance criteria into it.

When a proposed edit would choose a new audience, problem, promise, core loop,
product boundary, or other product intent, leave the file unchanged and ask the
user to settle that meaning. When the user wants to reopen the product
premise, invoke `define-product` with the current `PRODUCT.md`, the exact
ambiguity found, and the surfaces that depend on it, so it revises only that
part of the premise; it hands back the updated `PRODUCT.md`, after which this
pass resumes and reconciles the remaining durable surfaces against it.

## Maintain decision contracts

List `docs/decisions/` and read its index when it exists. Check for files
missing from the index, then compare relevant subjects with current context and
remaining specs.

Keep one editable decision file per subject. Use Git for history instead of
keeping old decision files in the active folder. When several files clearly
agree on one decision, merge them under the clearest stable subject name. Update
every inbound link before deleting redundant files.

Every decision contract created or edited in this pass needs a subject title
and these required sections:

- `Decisions`: every current rule;
- `Why`: the minimum reason needed to apply the rule without repeating the same
  debate.

Add these sections only when they carry current value:

- `Boundaries`: limits or exceptions;
- `Reconsider when`: a specific condition that should reopen the decision;
- `Still-rejected alternatives`: a future agent might retry a rejected path;
- `Evidence worth preserving`: a measurement or experiment would be costly to
  reproduce.

Remove status fields, supersession chains, adoption or update dates,
pull-request order, and old implementation notes. Keep a date when it sets a
current rule or tells future work when to reconsider.

Before removing other content, ask whether its absence could make a capable
future agent repeat the same proposal, investigation, experiment, or failed
approach under the same conditions. If so, keep the shortest explanation that
prevents repetition. Remove duplicates and alternatives that cannot reasonably
recur.

## Maintain related files

- In `docs/decisions/README.md`, keep exactly one entry per decision file:
  `- [subject](subject.md) — Read when ...`. Make every link work. The routing
  phrase says when to open the file, not what it decided.
- In `GLOSSARY.md`, keep only terms the project currently uses and rewrite a
  definition only when the project's language has already changed.
- For each `docs/specs/<slug>/` confirmed as shipped, first preserve every
  qualifying reusable decision in its owning contract, then delete the folder.
  If shipment is unclear, leave it and ask. In active specs, update only links
  and terms already confirmed elsewhere.
- In always-loaded instructions, keep repository working rules and a route to
  the decision index. Remove copied decision or product content without changing
  unrelated instructions.

Create, rename, or rebuild decision contracts and their index when the cleanup
requires it. Do not merge unrelated subjects or remove necessary instructions
to hit a size target. Report an index or always-loaded file that remains hard to
navigate and explain why.

## Reconcile the issue-tracker mirror

When `AGENTS.md` or `CLAUDE.md` carries an `## Issue tracker` section, read both
files and use the first section found and its linked detailed convention when
present. Existing inline conventions remain valid; the setup skill need not be

From toy-crane/skills, MIT.

More from toy-crane/skills

Signals

Stars
3
Forks
0
Open issues
0
Last commit (this folder)
Oct 9, 2026 (21 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from toy-crane

Done for you

Rather not set it up yourself?

Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.

See how we build AI agents →Read the case studies

Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.

launches shipped
50+
launches shipped
media placements
500+
media placements
founders on every account
2
founders on every account

Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block

Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.

Get maintain-project-context running in your business

Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.

Goes to the Chalk Labs founders, not the newsletter. Privacy

Need help deploying maintain-project-context?

CallGet help
Cite this page

maintain-project-context skill. Agent Mart. https://agentmart.dev/skills/toy-crane-maintain-project-context (last updated 2026-10-24)